feat(wizard): add floating page bot verification

Co-authored-by: Chan <dadachann@users.noreply.github.com>

.github/workflows/deploy-prod.yml

@@ -0,0 +1,59 @@
+name: Build and deploy production
+
+on:
+  push:
+    branches: [deploy/prod]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: langbot-production
+  cancel-in-progress: false
+
+env:
+  CORE_IMAGE: ${{ secrets.DOCKER_USERNAME }}/langbot
+  CLOUD_IMAGE: ${{ secrets.DOCKER_USERNAME }}/langbot-cloud-core
+  SPACE_REF: 58253c53933f95d81b035fbe2efedb55b6c1a82b
+
+jobs:
+  build-and-deploy:
+    runs-on: ubuntu-latest
+    environment: production
+    steps:
+      - uses: actions/checkout@v4
+      - uses: docker/setup-buildx-action@v3
+      - uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKER_USERNAME }}
+          password: ${{ secrets.DOCKER_PASSWORD }}
+      - name: Build exact Core image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          push: true
+          tags: |
+            ${{ env.CORE_IMAGE }}:prod-${{ github.sha }}
+            ${{ env.CORE_IMAGE }}:deploy-prod
+          cache-from: type=gha,scope=core-prod
+          cache-to: type=gha,mode=max,scope=core-prod
+      - name: Checkout production Cloud adapter
+        uses: actions/checkout@v4
+        with:
+          repository: langbot-app/langbot-space
+          ref: ${{ env.SPACE_REF }}
+          token: ${{ secrets.CLA_PAT }}
+          path: .space
+      - name: Build exact Cloud Core image
+        uses: docker/build-push-action@v6
+        with:
+          context: .space
+          file: .space/Dockerfile.cloud
+          push: true
+          build-args: LANGBOT_CORE_IMAGE=${{ env.CORE_IMAGE }}:prod-${{ github.sha }}
+          tags: |
+            ${{ env.CLOUD_IMAGE }}:prod-${{ github.sha }}
+            ${{ env.CLOUD_IMAGE }}:deploy-prod
+          cache-from: type=gha,scope=cloud-core-prod
+          cache-to: type=gha,mode=max,scope=cloud-core-prod

deploy/prod/deploy.sh

@@ -0,0 +1,97 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+cd /opt/langbot-cloud-prod
+TAG=${1:?usage: deploy.sh prod-<40-char-sha>}
+[[ "$TAG" =~ ^prod-[0-9a-f]{40}$ ]] || { echo 'invalid immutable image tag' >&2; exit 2; }
+[[ -s .env ]] || { echo '/opt/langbot-cloud-prod/.env is missing' >&2; exit 3; }
+
+rendered_compose=$(docker compose config)
+grep -Fq 'LANGBOT_SPACE_CONTROL_PLANE_URL: https://space.langbot.app' <<<"$rendered_compose" || {
+  echo 'Cloud control-plane URL must be https://space.langbot.app' >&2
+  exit 4
+}
+grep -Fq 'SPACE__URL: https://space.langbot.app' <<<"$rendered_compose" || {
+  echo 'Cloud user-facing Space URL must be https://space.langbot.app' >&2
+  exit 5
+}
+grep -Eq 'LANGBOT_TELEMETRY_INGEST_TOKEN: .+' <<<"$rendered_compose" || {
+  echo 'Cloud telemetry ingest token must be configured' >&2
+  exit 6
+}
+
+update_env() {
+  local key=$1 value=$2
+  python3 - "$key" "$value" <<'PY'
+from pathlib import Path
+import os
+import sys
+
+path = Path('.env')
+key, value = sys.argv[1:]
+lines = path.read_text().splitlines()
+updated = False
+for index, line in enumerate(lines):
+    if line.startswith(f'{key}='):
+        lines[index] = f'{key}={value}'
+        updated = True
+        break
+if not updated:
+    lines.append(f'{key}={value}')
+temporary = Path('.env.tmp')
+temporary.write_text('\n'.join(lines) + '\n')
+os.chmod(temporary, 0o600)
+temporary.replace(path)
+PY
+}
+update_env LANGBOT_IMAGE_TAG "$TAG"
+set -a
+. ./.env
+set +a
+: "${CLOUD_V2_CONTROL_PLANE_TOKEN:?CLOUD_V2_CONTROL_PLANE_TOKEN is required}"
+
+for attempt in 1 2 3 4 5; do
+  if docker compose pull postgres redis migrate plugin-runtime core; then
+    break
+  fi
+  if [ "$attempt" -eq 5 ]; then
+    echo "docker compose pull failed after $attempt attempts" >&2
+    exit 1
+  fi
+  delay=$((attempt * 10))
+  echo "docker compose pull failed (attempt $attempt/5); retrying in ${delay}s" >&2
+  sleep "$delay"
+done
+docker compose up -d postgres redis
+for _ in $(seq 1 60); do
+  if docker compose exec -T postgres pg_isready -U langbot_operator -d langbot >/dev/null 2>&1; then break; fi
+  sleep 2
+done
+docker compose exec -T postgres pg_isready -U langbot_operator -d langbot >/dev/null
+
+docker compose exec -T postgres psql -v ON_ERROR_STOP=1 -U langbot_operator -d langbot \
+  -v runtime_password="$POSTGRES_RUNTIME_PASSWORD" <<'SQL'
+SELECT format('CREATE ROLE langbot_runtime LOGIN PASSWORD %L', :'runtime_password')
+WHERE NOT EXISTS (SELECT 1 FROM pg_roles WHERE rolname = 'langbot_runtime')\gexec
+ALTER ROLE langbot_runtime PASSWORD :'runtime_password';
+GRANT CONNECT ON DATABASE langbot TO langbot_runtime;
+REVOKE CREATE ON SCHEMA public FROM PUBLIC, langbot_runtime;
+REVOKE ALL PRIVILEGES ON ALL TABLES IN SCHEMA public FROM langbot_runtime;
+REVOKE ALL PRIVILEGES ON ALL SEQUENCES IN SCHEMA public FROM langbot_runtime;
+ALTER DEFAULT PRIVILEGES FOR ROLE langbot_operator IN SCHEMA public REVOKE ALL ON TABLES FROM langbot_runtime;
+ALTER DEFAULT PRIVILEGES FOR ROLE langbot_operator IN SCHEMA public REVOKE ALL ON SEQUENCES FROM langbot_runtime;
+GRANT USAGE ON SCHEMA public TO langbot_runtime;
+SQL
+
+docker compose --profile tools run --rm migrate
+
+docker compose up -d --remove-orphans plugin-runtime core
+for _ in $(seq 1 90); do
+  if docker compose exec -T core python -c 'import urllib.request; urllib.request.urlopen("http://127.0.0.1:5300/healthz", timeout=3)' >/dev/null 2>&1; then
+    docker compose ps
+    exit 0
+  fi
+  sleep 2
+done
+docker compose logs --tail=200 core plugin-runtime >&2
+exit 1

deploy/prod/docker-compose.yml

@@ -0,0 +1,162 @@
+services:
+  postgres:
+    image: pgvector/pgvector:pg17
+    container_name: langbot-cloud-postgres
+    restart: unless-stopped
+    environment:
+      POSTGRES_DB: langbot
+      POSTGRES_USER: langbot_operator
+      POSTGRES_PASSWORD: ${POSTGRES_OPERATOR_PASSWORD}
+    volumes:
+      - postgres-data:/var/lib/postgresql/data
+    healthcheck:
+      test: [CMD-SHELL, "pg_isready -U langbot_operator -d langbot"]
+      interval: 5s
+      timeout: 5s
+      retries: 30
+    networks: [internal]
+
+  redis:
+    image: redis:7.4-alpine
+    container_name: langbot-cloud-redis
+    restart: unless-stopped
+    command: [redis-server, --appendonly, "yes", --requirepass, "${REDIS_PASSWORD}"]
+    volumes:
+      - redis-data:/data
+    healthcheck:
+      test: [CMD-SHELL, "redis-cli -a \"$${REDIS_PASSWORD}\" ping | grep PONG"]
+      interval: 5s
+      timeout: 5s
+      retries: 20
+    environment:
+      REDIS_PASSWORD: ${REDIS_PASSWORD}
+    networks: [internal]
+
+  migrate:
+    image: rockchin/langbot-cloud-core:${LANGBOT_IMAGE_TAG}
+    profiles: [tools]
+    command: [uv, run, langbot, migrate, --cloud]
+    environment: &core-env
+      TZ: Asia/Shanghai
+      SYSTEM__INSTANCE_ID: ${CLOUD_V2_INSTANCE_UUID}
+      SYSTEM__EDITION: cloud
+      SYSTEM__RECOVERY_KEY: ${SYSTEM_RECOVERY_KEY}
+      SYSTEM__JWT__SECRET: ${JWT_SECRET}
+      SYSTEM__LIMITATION__MAX_BOTS: "2"
+      SYSTEM__LIMITATION__MAX_PIPELINES: "3"
+      SYSTEM__LIMITATION__MAX_EXTENSIONS: "3"
+      SYSTEM__LIMITATION__MAX_KNOWLEDGE_BASES: "2"
+      API__WEBHOOK_PREFIX: https://cloud.langbot.app
+      API__WEBUI_URL: https://cloud.langbot.app
+      WORKSPACE__INVITATIONS__PUBLIC_WEB_URL: https://cloud.langbot.app
+      DATABASE__USE: postgresql
+      DATABASE__POSTGRESQL__URL: postgresql+asyncpg://langbot_runtime:${POSTGRES_RUNTIME_PASSWORD}@postgres:5432/langbot
+      DATABASE__CLOUD_MIGRATION__OPERATOR_DSN_ENV: LANGBOT_CLOUD_MIGRATION_DSN
+      LANGBOT_CLOUD_MIGRATION_DSN: postgresql://langbot_operator:${POSTGRES_OPERATOR_PASSWORD}@postgres:5432/langbot
+      VDB__USE: pgvector
+      VDB__PGVECTOR__USE_BUSINESS_DATABASE: "true"
+      VDB__PGVECTOR__ALLOWED_DIMENSIONS: "384,512,768,1024,1536"
+      PLUGIN__ENABLE: "true"
+      PLUGIN__RUNTIME_WS_URL: ws://plugin-runtime:5400/control/ws
+      PLUGIN__DISPLAY_PLUGIN_DEBUG_URL: wss://cloud.langbot.app/plugin/debug/ws
+      PLUGIN__WORKER__MAX_CPUS: "0.25"
+      PLUGIN__WORKER__MAX_MEMORY_MB: "256"
+      PLUGIN__WORKER__MAX_PIDS: "128"
+      PLUGIN__WORKER__MAX_WORKERS: "16"
+      PLUGIN__WORKER__MAX_TOTAL_CPUS: "4.0"
+      PLUGIN__WORKER__MAX_TOTAL_MEMORY_MB: "4096"
+      PLUGIN__WORKER__REQUIRE_HARD_LIMITS: "true"
+      LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN: ${PLUGIN_RUNTIME_CONTROL_TOKEN}
+      # Cloud v2 currently grants no managed Box capability. Keep the shared
+      # runtime deployed but disable Core integration until a hard-quota-capable
+      # backend can satisfy the fail-closed Cloud readiness contract.
+      BOX__ENABLED: "false"
+      BOX__BACKEND: nsjail
+      BOX__RUNTIME__ENDPOINT: ws://box:5410
+      BOX__ADMISSION__REQUIRED: "true"
+      BOX__ADMISSION__LOGICAL_SESSION_ID: global
+      BOX__ADMISSION__REQUIRED_BACKEND: nsjail
+      BOX__ADMISSION__MAX_SESSIONS: "1"
+      BOX__ADMISSION__MAX_MANAGED_PROCESSES: "0"
+      BOX__ADMISSION__CPUS: "0.25"
+      BOX__ADMISSION__MEMORY_MB: "256"
+      BOX__ADMISSION__WORKSPACE_QUOTA_MB: "256"
+      BOX__LOCAL__HOST_ROOT: /app/data/box
+      BOX__LOCAL__DEFAULT_WORKSPACE: /app/data/box
+      BOX__LOCAL__ALLOWED_MOUNT_ROOTS: /app/data/box
+      LANGBOT_BOX_CONTROL_TOKEN: ${BOX_CONTROL_TOKEN}
+      MCP__STDIO__ENABLED: "false"
+      LANGBOT_SPACE_CONTROL_PLANE_URL: https://space.langbot.app
+      LANGBOT_SPACE_CONTROL_PLANE_TOKEN: ${CLOUD_V2_CONTROL_PLANE_TOKEN}
+      LANGBOT_TELEMETRY_INGEST_TOKEN: ${CLOUD_V2_CONTROL_PLANE_TOKEN}
+      LANGBOT_SPACE_CONTROL_PLANE_PUBLIC_KEY: ${CLOUD_V2_MANIFEST_PUBLIC_KEY}
+      LANGBOT_SPACE_CONTROL_PLANE_KEY_ID: ${CLOUD_V2_MANIFEST_KEY_ID}
+      SPACE__URL: https://space.langbot.app
+    depends_on:
+      postgres: {condition: service_healthy}
+    networks: [internal]
+
+  plugin-runtime:
+    image: rockchin/langbot:${LANGBOT_IMAGE_TAG}
+    container_name: langbot-cloud-plugin-runtime
+    restart: unless-stopped
+    command: [uv, run, python, -m, langbot_plugin.cli.__init__, rt]
+    environment:
+      LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN: ${PLUGIN_RUNTIME_CONTROL_TOKEN}
+    volumes:
+      - plugin-data:/app/data
+      - /sys/fs/cgroup:/sys/fs/cgroup:rw
+    cgroup: host
+    privileged: true
+    expose: ["5400"]
+    networks: [internal]
+
+  box:
+    image: rockchin/langbot:${LANGBOT_IMAGE_TAG}
+    container_name: langbot-cloud-box
+    restart: unless-stopped
+    command: [uv, run, lbp, box, --host, 0.0.0.0, --ws-control-port, "5410"]
+    environment:
+      LANGBOT_BOX_CONTROL_TOKEN: ${BOX_CONTROL_TOKEN}
+      LANGBOT_BOX_ROOT: /app/data/box
+    volumes:
+      - box-data:/app/data/box
+      - /sys/fs/cgroup:/sys/fs/cgroup:rw
+    cgroup: host
+    privileged: true
+    expose: ["5410"]
+    networks: [internal]
+
+  core:
+    image: rockchin/langbot-cloud-core:${LANGBOT_IMAGE_TAG}
+    container_name: langbot-cloud-core
+    restart: unless-stopped
+    environment: *core-env
+    volumes:
+      - core-data:/app/data
+      - box-data:/app/data/box
+    depends_on:
+      postgres: {condition: service_healthy}
+      redis: {condition: service_healthy}
+      plugin-runtime: {condition: service_started}
+      box: {condition: service_started}
+    expose: ["5300"]
+    healthcheck:
+      test: [CMD-SHELL, "python -c 'import urllib.request; urllib.request.urlopen(\"http://127.0.0.1:5300/healthz\", timeout=3)'" ]
+      interval: 10s
+      timeout: 5s
+      retries: 30
+      start_period: 30s
+    networks: [internal, shared-network]
+
+networks:
+  internal:
+  shared-network:
+    external: true
+
+volumes:
+  postgres-data:
+  redis-data:
+  plugin-data:
+  box-data:
+  core-data:

docker/docker-compose.yaml

@@ -14,8 +14,8 @@ services:
     restart: on-failure
     environment:
       - TZ=Asia/Shanghai
-      # Shared with the langbot service and sent only as a WebSocket handshake
-      # header. Generate with: openssl rand -hex 32
+      # Optional. Leave unset on both OSS services, or set the same value on
+      # both to protect the control WebSocket. Generate with: openssl rand -hex 32
       - LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN=${LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN:-}
       # Process-wide admission for every asyncio.to_thread() call.
       - LANGBOT_BLOCKING_EXECUTOR_MAX_WORKERS=${LANGBOT_BLOCKING_EXECUTOR_MAX_WORKERS:-8}
@@ -77,8 +77,7 @@ services:
     restart: on-failure
     environment:
       - TZ=Asia/Shanghai
-      # Must match langbot_plugin_runtime. Empty/missing values make the
-      # external control channel fail closed.
+      # Optional. Leave unset on both OSS services, or match plugin Runtime.
       - LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN=${LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN:-}
       # Must match the value supplied to langbot_box. The token is sent only
       # in WebSocket handshake headers, never in URLs or action payloads.

docs/review/model-reasoning-control-design.md

@@ -0,0 +1,476 @@
+# 模型思考控制设计方案
+
+> 日期：2026-07-31
+> 状态：Phase 1 已审核并实现
+> 范围：LangBot 主仓库的模型配置、LiteLLM 请求层、Local Agent、Web 管理面板、监控与测试
+
+## 1. 结论
+
+建议为 LangBot 增加一套与厂商参数解耦的“思考策略”模型，并明确区分三个概念：
+
+1. **思考能力**：模型是否支持思考，以及支持开关、档位还是 token 预算。
+2. **思考策略**：一次请求选择厂商默认、关闭、开启或指定思考档位。
+3. **思考展示**：是否把模型返回的思考内容展示给最终用户。
+
+现有 `remove-think` 只属于第 3 类。它会过滤输出，但不会阻止模型思考，也不会降低思考 token、费用或延迟。新能力不应复用或改写这个字段。
+
+推荐实现原则：
+
+- 默认值为 `provider_default`，不向上游增加任何新参数，现有模型行为完全不变。
+- 用户显式选择的策略必须被准确执行；无法准确执行时返回明确错误，不静默降级。
+- LangBot 内部只保存统一策略，Provider 请求层负责翻译成各厂商参数。
+- `extra_args` 保留为高级逃生口，但不能成为主 UI 的思考配置方式。
+- 模型页只管理并展示能力；可写策略归属于 Local Agent 流水线，同一模型可在不同业务中使用不同思考量。
+- 原始 reasoning 数据与展示文本分开保存，保证多轮对话、工具调用和签名字段不丢失。
+
+## 2. 调研结论
+
+### 2.1 可验证资料
+
+本次结论基于以下可验证来源：
+
+- OpenAI 官方 Reasoning Guide：`reasoning.effort` 的可选值由模型决定，可包括 `none`、`minimal`、`low`、`medium`、`high`、`xhigh`、`max`；低档位偏向低延迟和低 token，高档位偏向质量。
+  - https://developers.openai.com/api/docs/guides/reasoning#reasoning-effort
+- LangBot 锁定的 LiteLLM `1.88.1` 实现。`uv.lock` 已锁定该版本，本地缓存中的适配代码可以确认 LangBot 实际依赖所支持的翻译行为。
+- LangBot 当前实现：模型级 `extra_args` 会在 `LiteLLMRequester._build_completion_args()` 中直接合并到 `acompletion()` 参数。
+
+Anthropic、Google 和 LiteLLM 的官方文档域名在本次环境中被浏览器策略禁止访问，因此下表中这些厂商的结论以 LiteLLM `1.88.1` 实际适配代码为准。实施前应再用对应厂商官方文档做一次参数范围核验，尤其是模型代际和允许值。
+
+### 2.2 厂商差异矩阵
+
+| Provider / 生态 | 可控制能力 | LiteLLM 1.88.1 统一入口 | 关键限制 | 建议支持级别 |
+| --- | --- | --- | --- | --- |
+| OpenAI | 思考档位，部分模型支持 `none` | `reasoning_effort` | 每个模型支持的档位不同，不能把 `none` 当成通用能力 | 首批完整支持 |
+| Anthropic | 旧模型使用 extended thinking + token budget；新模型可用 adaptive thinking + effort | `reasoning_effort` 或 `thinking` | `none` 表示不发送 thinking；新旧模型的映射不同 | 首批完整支持 |
+| Gemini | 2.x 主要映射为 `thinkingBudget`；3.x 主要映射为 `thinkingLevel` | `reasoning_effort` 或 `thinking` | Gemini 3 的 `none` 可能只能降到最低档，不能保证真正关闭 | 首批支持，但严格限制关闭语义 |
+| DeepSeek | 开启/关闭；当前适配不支持预算档位 | `thinking={type: enabled}`；非 `none` effort 会映射成开启 | 多轮思考模式要求回传 `reasoning_content` | 首批开关支持 |
+| xAI | 思考档位 | `reasoning_effort` | 仅 reasoning-capable 模型接受 | 首批完整支持 |
+| Ollama | `think` 布尔值；部分模型接受 low/medium/high | `reasoning_effort` | 非 gpt-oss 模型的档位可能退化为布尔开关 | 首批支持，按模型能力裁剪 UI |
+| OpenRouter | 聚合多厂商的 reasoning 参数 | `reasoning_effort`、`thinking` | 实际能力由路由后的模型决定 | 首批支持，能力未知时要求测试 |
+| Volcengine / Doubao | `thinking.type` 支持 enabled/disabled/auto | LiteLLM `volcengine` 适配器支持 `thinking` | LangBot 当前 manifest 使用 `openai`，不会进入该适配器 | 第二批，先修正路由并回归 |
+| Bailian / Qwen | 厂商兼容接口有独立思考开关/预算 | LiteLLM `dashscope` 适配器目前未提供统一 reasoning 映射 | LangBot 当前 manifest 使用 `openai`，只能通过高级参数透传 | 第二批，实施前核对官方字段 |
+| 其他 OpenAI-compatible 网关 | 取决于网关 | 尝试标准 `reasoning_effort` | 不能仅凭模型名推断完整能力 | 保守支持，默认不自动开启 |
+
+### 2.3 对 LangBot 的直接含义
+
+不能把这个功能实现成单一 `enable_thinking: bool`，原因如下：
+
+- 有的模型只有开关，有的模型只有档位，有的模型允许精确 token 预算。
+- 有的模型本身始终推理，只能降低思考量，无法真正关闭。
+- 同一个通用档位在不同厂商会映射成不同的实际预算。
+- 聚合网关和自定义 OpenAI-compatible 服务无法可靠地通过模型名识别能力。
+- “不展示思考内容”不等于“关闭思考”。
+
+## 3. 当前项目现状
+
+### 3.1 已有能力
+
+- `LLMModel.extra_args` 是 JSON 字段，Web 端已有通用高级参数编辑器。
+- `LiteLLMRequester` 会按“模型级 `extra_args`，再调用级 `extra_args`”的顺序合并参数。
+- LiteLLM 已统一处理多个 Provider 的 `reasoning_effort`、`thinking` 和返回的 `reasoning_content`。
+- `LocalAgentRunner` 的非流式、流式、工具调用和 fallback 路径都经过 `RuntimeProvider.invoke_llm*()`。
+- `remove-think` 已能控制 `<think>` 或独立 reasoning 内容是否进入展示文本。
+- Gemini 工具调用所需的 `provider_specific_fields` / thought signature 已有保留逻辑和单元测试。
+
+### 3.2 现有缺口
+
+- 管理员只能手写 `extra_args`，没有统一语义、能力提示和校验。
+- `remove-think` 名称容易被误解为关闭模型思考。
+- 模型扫描只识别 `vision` 和 `func_call`，没有 reasoning 能力。
+- 当前返回处理会把 `reasoning_content` 拼进 `<think>` 文本后删除原字段，可能损失多轮思考所需的结构化数据。
+- DeepSeek 思考模式需要在后续轮次回传 `reasoning_content`，当前链路不能保证完整保留。
+- Pipeline 只能选择模型，不能针对业务覆盖模型的思考策略。
+- 监控只记录总输入/输出 token，没有单独展示 reasoning token。
+- 部分 Provider manifest 仍声明为通用 `openai`，导致 LiteLLM 的厂商专用翻译器不会生效。
+
+### 3.3 预计改动地图
+
+| 层 | 主要文件 | 责任 |
+| --- | --- | --- |
+| 持久化 | `src/langbot/pkg/entity/persistence/model.py`、`src/langbot/pkg/persistence/alembic/versions/` | 新增 `reasoning_config` JSON 列和 Alembic 迁移 |
+| 模型服务 | `src/langbot/pkg/api/http/service/model.py` | CRUD 校验、冲突检测、测试模型时使用统一策略 |
+| HTTP 控制器 | `src/langbot/pkg/api/http/controller/groups/provider/models.py` | 继续复用现有模型路由，不新增平行 API |
+| 模型管理 | `src/langbot/pkg/provider/modelmgr/modelmgr.py` | 临时模型、数据库模型与扫描结果加载新字段 |
+| 请求抽象 | `src/langbot/pkg/provider/modelmgr/requester.py` | 定义能力查询和 reasoning 参数构建接口 |
+| LiteLLM 适配 | `src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py` | 能力识别、策略翻译、参数合并、reasoning 返回保留 |
+| Provider manifest | `src/langbot/pkg/provider/modelmgr/requesters/*.yaml` | 必要时修正 Provider 路由；相关变更放到独立阶段 |
+| Agent 调用 | `src/langbot/pkg/provider/runners/localagent.py` | 所有非流式、流式、工具调用、fallback 路径传递统一策略 |
+| Pipeline 元数据 | `src/langbot/templates/metadata/pipeline/ai.yaml` | 第二阶段加入 Pipeline 级覆盖 |
+| 输出配置 | `src/langbot/templates/metadata/pipeline/output.yaml` | 保留键名，澄清 `remove-think` 只控制展示 |
+| Web 类型/API | `web/src/app/infra/entities/api/index.ts`、`web/src/app/infra/http/BackendClient.ts` | 增加配置与能力响应类型 |
+| 模型 UI | `web/src/app/home/components/models-dialog/` | 能力标记、策略控件、校验、模型测试 |
+| i18n | `web/src/i18n/locales/` | 至少补齐英文、简体中文及项目已有覆盖语言 |
+| 测试 | `tests/unit_tests/provider/`、`web/tests/` | 翻译、服务、流式 round-trip、前端状态测试 |
+
+Phase 1 不修改 `langbot-plugin-sdk` 的公共实体或运行时协议。现有 `provider_message.Message.provider_specific_fields` 已可承载 Provider 原始 reasoning 数据；只有后续要把 reasoning 升级为跨插件公开实体时，才需要跨仓库 SDK 变更。
+
+## 4. 领域模型
+
+### 4.1 统一策略
+
+新增 `ReasoningConfig`，保存于 LLM 模型，Pipeline 可提供同结构覆盖。产品层只暴露一个离散档位：
+
+```json
+{
+  "level": "provider_default"
+}
+```
+
+字段定义：
+
+| 字段 | 类型 | 含义 |
+| --- | --- | --- |
+| `level` | `provider_default \| disabled \| enabled \| minimal \| low \| medium \| high \| xhigh \| max` | 同时表达开关和思考强度 |
+
+校验规则：
+
+- `provider_default`：不发送任何 reasoning 参数，保持厂商和模型默认行为。
+- `disabled`：明确关闭；仅当模型可真正关闭时允许保存/运行。
+- `enabled`：明确开启，但由 Provider 决定具体强度，适用于只有开关的模型。
+- `minimal` 到 `max`：明确开启，并指定强度；仅允许选择模型实际支持的档位。
+- 厂商的 `auto` 统一映射为 `provider_default`，不再增加一个重复状态。
+- 精确 token 预算不进入主数据结构。少数需要预算的场景继续通过高级参数配置，并由模型测试接口校验。
+
+### 4.2 能力描述
+
+沿用现有 `LLMModel.abilities`，新增 `reasoning` 能力标记。同时由后端在 API 返回中计算只读的 `reasoning_capabilities`：
+
+```json
+{
+  "supported": true,
+  "controls": ["toggle", "effort"],
+  "efforts": ["none", "low", "medium", "high"],
+  "can_disable": true,
+  "source": "litellm"
+}
+```
+
+设计约束：
+
+- `abilities` 仍是用户可编辑的粗粒度能力，符合现有 `vision`、`func_call` 模式。
+- `reasoning_capabilities` 不持久化，优先从 LiteLLM 模型元数据计算，避免模型升级后数据库残留过期能力。
+- 无法识别的自定义模型返回 `supported: null`、`source: unknown`，不猜测。
+- 用户可手动添加 `reasoning` ability，但未知能力模型必须先通过“测试模型”验证显式策略。
+- UI 只展示后端声明可用的控件；未知模型保留 Provider Default 和高级参数入口。
+
+### 4.3 持久化
+
+在 `llm_models` 表新增 JSON 列：
+
+```text
+reasoning_config JSON NOT NULL DEFAULT {"level":"provider_default"}
+```
+
+使用 Alembic 新迁移，不修改冻结的 legacy migration。
+
+该列作为已实现版本的兼容字段保留；新的模型页不再提供写入口，Local Agent 请求以流水线中按模型 UUID 保存的策略为准。
+
+不建议把内部策略塞进 `extra_args`，原因是当前 `extra_args` 会原样发送给 LiteLLM；使用保留键会让内部元数据泄漏到上游，并使高级参数与产品配置难以区分。
+
+## 5. 配置优先级与请求流程
+
+### 5.1 优先级
+
+```text
+Pipeline 当前候选模型策略
+        ↓ 缺少配置时固定为 provider_default
+Provider / 模型默认行为
+```
+
+请求参数合并顺序：
+
+```text
+基础参数
+  -> 模型 extra_args
+  -> 调用级 extra_args
+  -> 统一 reasoning 策略翻译结果（最后应用）
+```
+
+统一策略最后应用，可以确保流水线行为不受模型页历史设置影响。为了避免用户困惑，保存和测试时要检测 `extra_args` 中的冲突字段；当 `level != provider_default` 时，发现以下字段应直接报错：
+
+- `reasoning_effort`
+- `thinking`
+- `reasoning`
+- `extra_body` 内已知的 `thinking`、`enable_thinking`、`thinking_budget` 等字段
+
+当 `level == provider_default` 时继续允许这些高级参数，保证旧配置兼容。
+
+### 5.2 翻译层
+
+在 `pkg/provider/modelmgr/` 内新增独立的 reasoning 规范化模块，职责是：
+
+1. 读取当前流水线候选模型的请求级策略。
+2. 查询 `ProviderAPIRequester.get_reasoning_capabilities(model)`。
+3. 严格校验策略是否可以准确执行。
+4. 生成 LiteLLM 参数，不直接发 HTTP。
+5. 返回可观测的“最终生效策略”供日志和测试使用。
+
+建议接口：
+
+```python
+class ProviderAPIRequester:
+    def get_reasoning_capabilities(self, model: RuntimeLLMModel) -> ReasoningCapabilities: ...
+
+    def build_reasoning_args(
+        self,
+        model: RuntimeLLMModel,
+        config: ReasoningConfig,
+    ) -> dict[str, Any]: ...
+```
+
+LiteLLMRequester 默认优先生成统一参数：
+
+- 强度档位：`reasoning_effort=<level>`
+- 仅开启：`thinking={"type":"enabled"}` 或 Provider 等价参数
+- 关闭：优先 `reasoning_effort="none"`
+- 高级参数中的精确预算：`thinking={"type":"enabled","budget_tokens":N}`
+
+Provider 特例只放在 requester 翻译层，不进入 Pipeline 或平台适配器。
+
+### 5.3 Provider 特例
+
+- **Gemini 3**：如果 LiteLLM 能力表不能确认真正关闭，`disabled` 必须报“不支持关闭，可选择 Provider Default 或最低档”，不能把 `none` 静默映射成 low/minimal。
+- **DeepSeek**：所有非 `none` 档位最终都只是开启。能力 API 只返回 `toggle`，UI 不显示档位；多轮必须保存并回传 `reasoning_content`。
+- **Ollama**：仅对明确支持等级的模型展示 effort；其他模型只展示开关。
+- **OpenRouter**：以路由后的模型能力为准。模型未知时允许 Provider Default，显式策略必须通过测试接口。
+- **Volcengine**：使用 `thinking.type=enabled/disabled/auto`。应先让该 requester 进入 LiteLLM `volcengine` 适配器，或增加等价的明确翻译，不能依赖模型名。
+- **Bailian/Qwen**：作为第二批 Provider 专用翻译。实施前核对官方字段、模型范围、预算上下限和流式返回结构，不凭经验写接口。
+
+## 6. 返回数据与思考展示
+
+### 6.1 保留原始 reasoning
+
+当前 `LiteLLMRequester` 会读取 `reasoning_content`，将其拼接成 `<think>` 文本，再删除原字段。建议改为：
+
+```text
+上游 reasoning_content
+  ├─ 原样保存在 Message.provider_specific_fields.reasoning_content
+  └─ 根据 remove-think 决定是否渲染为 <think>...</think>
+```
+
+流式路径需要在 accumulator 中分别累计 `content` 与 `reasoning_content`，最终消息必须携带结构化 reasoning。不能只依赖已经渲染的 `<think>` 文本反向解析。
+
+这样可以同时满足：
+
+- `remove-think=true` 时用户看不到思考内容，但多轮协议仍能回传必要数据。
+- `remove-think=false` 时保持当前用户体验。
+- DeepSeek 多轮 thinking 不丢上下文。
+- Gemini thought signature、Anthropic thinking block 等 Provider 字段可以继续按结构化方式 round-trip。
+
+### 6.2 现有字段处理
+
+保留数据库和 Pipeline 配置键 `remove-think`，避免破坏兼容。Web 文案改为更准确的：
+
+- 中文：`向用户展示思考过程`
+- 英文：`Show reasoning process`
+
+UI 使用正向开关，保存时转换回 `remove-think = !showReasoning`。文案必须强调它只影响展示，不影响模型是否思考、token 或费用。
+
+## 7. Web 管理面板
+
+### 7.1 模型编辑
+
+模型页只承担能力管理和只读展示：
+
+1. `Reasoning` ability 复选框与 Vision、Function Calling 并列，供无法自动识别的自定义模型手动声明能力。
+2. 模型卡片使用简短图标或 badge 标识 reasoning 能力。
+3. 模型页不提供可写思考挡位，避免模型默认值与流水线策略形成两个控制源。
+
+### 7.2 Local Agent 流水线策略
+
+在 Local Agent 的主模型和每一个 fallback 模型下分别显示紧凑离散滑杆：
+
+1. `Provider 默认` 始终为首个选项；选择它时不向上游增加任何思考参数。
+2. 完整档位顺序为：`Provider 默认 / 关闭 / 开启 / 最低 / 低 / 中 / 高 / 极高 / 最大`。
+3. 前端只渲染后端为该模型返回的可用档位；仅开关模型显示 `Provider 默认 / 关闭 / 开启`。
+4. 模型不能真正关闭时不提供 `关闭`；能力未知时只显示不可调的 `Provider 默认`。
+5. 主模型和 fallback 分别保存策略，切换候选模型时不会把一个模型的挡位错误应用到另一个模型。
+6. Dify、Coze、Langflow、n8n 等外部 Runner 不显示该控件，因为 LangBot 不直接发起其内部模型请求。
+
+流水线配置保持旧格式兼容，并在模型选择对象中增加按 UUID 保存的映射：
+
+```json
+{
+  "model": {
+    "primary": "primary-model-uuid",
+    "fallbacks": ["fallback-model-uuid"],
+    "reasoning": {
+      "primary-model-uuid": "high"
+    }
+  }
+}
+```
+
+`provider_default` 不写入映射；缺少 `reasoning` 的旧流水线天然等价于全部使用 Provider 默认。
+
+滑杆交互要求：轨道使用现有主色和中性灰，不使用渐变；当前档位同时显示文字；支持键盘方向键和正确的 ARIA value text；窄屏下不溢出。
+
+### 7.3 i18n
+
+新增文案至少覆盖 `en_US`、`zh_Hans`；`ja_JP` 在模型面板现有同类字段已覆盖时同步补齐。不要把厂商参数名直接作为用户文案。
+
+## 8. API、MCP 与 Skill
+
+### 8.1 HTTP API
+
+模型 CRUD 增加：
+
+- 请求字段：`reasoning_config`
+- 响应字段：`reasoning_config`
+- 只读字段：`reasoning_capabilities`
+
+模型测试接口必须使用与真实请求完全相同的规范化和翻译逻辑，并在失败时返回可操作错误，例如：
+
+```text
+Model gemini-3-... cannot disable reasoning.
+Supported controls: effort=[low, medium, high].
+```
+
+可选增加只读调试信息，仅在测试接口返回：
+
+```json
+{
+  "effective_reasoning": {
+    "level": "low",
+    "translated_keys": ["reasoning_effort"]
+  }
+}
+```
+
+不得返回 API key、完整请求正文或原始思考内容。
+
+### 8.2 MCP 与技能
+
+当前 MCP 仅列出模型 Provider，没有完整模型 CRUD 工具。如果本次不新增 agent-accessible HTTP 操作，则无需强行新增 MCP 工具。
+
+如果后续让 Agent 修改模型思考策略，则必须同一提交更新：
+
+- `src/langbot/pkg/api/mcp/server.py`
+- 对应的 `skills/` 文档
+- 参数 schema 和安全说明
+
+## 9. 监控与可观测性
+
+控制思考量后，管理员需要判断质量、延迟和成本是否值得。建议第二阶段增加：
+
+- `reasoning_tokens`：从 `completion_tokens_details.reasoning_tokens` 或 Provider 等价字段提取。
+- `effective_reasoning_level`：记录规范化后的生效档位，不记录原始思考内容。
+- 模型监控页展示输入 token、可见输出 token、reasoning token、总延迟。
+- Provider 不返回细分 token 时显示未知，不推算。
+
+安全要求：日志、监控、debug API 默认都不得记录 reasoning 原文。思考内容可能包含敏感信息或系统提示，不应因为新增配置而扩大持久化范围。
+
+## 10. 兼容与迁移
+
+### 10.1 数据迁移
+
+- 所有现有 LLM 记录迁移为 `{"level":"provider_default"}`。
+- 不自动解析或迁移现有 `extra_args` 中的 reasoning 参数，避免误判嵌套结构和 Provider 语义。
+- UI 检测到旧 `extra_args` reasoning 字段时显示“由高级参数控制”，统一策略保持 Provider Default。
+- 用户主动改成统一策略时，要求先移除冲突高级参数。
+
+### 10.2 运行时兼容
+
+- `provider_default` 不产生任何新增请求参数。
+- 不改变现有 `remove-think` 的存储键和默认值。
+- 不改变已有 Provider 的 `litellm_provider`，除非该 Provider 在专项回归后单独切换。
+- `drop_params` 不能用于掩盖显式 reasoning 配置错误；显式策略被丢弃应视为失败。
+- 自托管和 toB 环境中的自定义兼容接口保持可用，未知能力不阻止 Provider Default 请求。
+
+## 11. 实施拆分
+
+### Phase 1：统一基础设施与主流 Provider
+
+- Alembic 增加 `llm_models.reasoning_config`。
+- Backend 模型实体、CRUD、测试接口支持统一配置。
+- LiteLLMRequester 增加能力查询、严格校验和参数翻译。
+- 支持 OpenAI、Anthropic、Gemini、DeepSeek、xAI、Ollama、OpenRouter 的已验证 LiteLLM 路径。
+- 修复结构化 reasoning 的非流式/流式保留。
+- 模型面板增加 reasoning ability 与只读能力标识。
+- Local Agent 主模型和每个 fallback 增加独立的请求级策略。
+
+### Phase 2：国内 Provider
+
+- 专项核对并支持 Volcengine/Doubao、Bailian/Qwen。
+- 对相关 requester 的 `litellm_provider` 变更做独立回归，避免把 reasoning 功能和通用请求行为回归混在一起。
+- 补齐扫描结果中的 reasoning capability。
+
+### Phase 3：监控与评估
+
+- 持久化 reasoning token 和生效策略。
+- 监控页增加 reasoning 成本/延迟指标。
+- 建立不同 effort 的离线质量、首 token 延迟、总耗时和 token 对比基线。
+
+## 12. 测试方案
+
+### 12.1 单元测试
+
+- `ReasoningConfig` 所有合法/非法组合。
+- `provider_default` 不产生任何新增参数。
+- 显式配置覆盖模型/调用 `extra_args` 的顺序。
+- reasoning 配置与高级参数冲突时拒绝。
+- OpenAI 档位原样映射。
+- Anthropic 档位映射，以及高级参数预算兼容。
+- Gemini 2 budget、Gemini 3 level，以及不支持真正关闭时拒绝。
+- DeepSeek 只显示/接受 toggle，非 `none` effort 不伪装成不同档位。
+- Ollama 布尔与分级模型差异。
+- Volcengine enabled/disabled/auto 翻译。
+- 未知 Provider 只允许 Provider Default，或在显式测试后使用标准参数。
+- 非流式 `reasoning_content` 保存到 `provider_specific_fields`。
+- 流式 reasoning 分片累计后仍能 round-trip。
+- Gemini thought signature 和工具调用现有测试不能回归。
+
+### 12.2 服务与持久化测试
+
+- 新建、读取、更新模型的 `reasoning_config`。
+- Alembic 从当前 head 升级后默认值正确。
+- 模型测试接口与真实 Local Agent 使用同一翻译函数。
+- 旧模型、旧 `extra_args` 和 `remove-think` 行为不变。
+
+### 12.3 前端测试
+
+- 能力不同的模型显示正确控件。
+- 离散滑杆只能停在后端返回的可用档位。
+- 当前档位文字、键盘操作和 ARIA value text 正确。
+- 仅开关模型、不可关闭模型、完整档位模型分别显示正确刻度。
+- fallback 能力不兼容时阻止保存并给出明确提示。
+- 中英文文案完整，移动端 Popover 不溢出。
+
+### 12.4 Provider 冒烟测试
+
+至少选取以下真实或可控 mock：
+
+- 一个支持 `none` 的 OpenAI reasoning 模型。
+- 一个不支持 `none` 的 reasoning 模型。
+- 一个 Anthropic adaptive thinking 模型。
+- 一个 Gemini 2.x 与一个 Gemini 3.x 模型。
+- 一个 DeepSeek hybrid thinking 模型，执行两轮含工具调用对话。
+- 一个 Ollama 本地 reasoning 模型。
+- 一个 OpenAI-compatible 自定义网关，验证 Provider Default 完全不变。
+
+每个模型比较 Provider Default、最低档、中档、高档或关闭，记录成功率、首 token 延迟、总耗时、总 token 和 reasoning token（若可用）。
+
+## 13. 风险与控制
+
+| 风险 | 影响 | 控制措施 |
+| --- | --- | --- |
+| 将“最低思考”误当成“关闭” | 用户以为节省了成本，实际仍在推理 | `can_disable` 严格校验，不静默降级 |
+| 模型能力表过期 | 新模型无法配置或旧模型报错 | 能力未知时保守；允许测试；升级 LiteLLM 时回归 |
+| 高 effort 导致延迟/费用陡增 | 用户体验和预算风险 | 默认 Provider Default；UI 提示；后续监控 reasoning token |
+| `extra_args` 与统一配置冲突 | 实际生效值不可预测 | 保存/测试时拒绝冲突；统一策略最后应用 |
+| reasoning 原文进入日志 | 敏感信息泄露 | 不记录原文，只记录策略和 token |
+| 多轮 reasoning 丢失 | 工具调用或后续轮次失败/降质 | 结构化保存并 round-trip；流式专项测试 |
+| 修改 Provider 路由造成通用回归 | 非 reasoning 请求也受影响 | 国内 Provider 路由放第二阶段，独立提交和回归 |
+
+## 14. 需要审核确认的决策
+
+1. **是否同意三层分离**：能力、策略、展示互不替代，保留 `remove-think` 仅控制展示。
+2. **是否同意严格语义**：显式关闭无法准确执行时直接报错，不自动降为最低思考。
+3. **是否同意请求级配置**：流水线按模型 UUID 保存挡位，不把产品配置塞进 `extra_args`。
+4. **是否同意 Runner 边界**：仅 Local Agent 展示控制项，外部 Runner 由其外部系统管理模型策略。
+5. **是否同意保守默认**：所有现有模型迁移为 Provider Default，不自动开启、关闭或迁移旧高级参数。
+6. **是否把结构化 reasoning 保留纳入第一阶段**：这是 DeepSeek 多轮和工具调用正确性的必要条件，建议必须纳入。
+
+## 15. 推荐审核结果
+
+建议按以上 6 项全部通过，并将 Phase 1 作为一个完整功能单元实施。不要只增加前端开关或只在 `extra_args` 中写 `reasoning_effort`；那样虽然改动小，但会继续混淆展示与推理、无法处理 Provider 差异，也无法保证多轮对话正确性。

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langbot"
-version = "4.10.6"
+version = "4.10.7"
 description = "Production-grade platform for building agentic IM bots"
 readme = "README.md"
 license-files = ["LICENSE"]
@@ -23,7 +23,7 @@ dependencies = [
     "pynacl>=1.5.0", # Required for Discord voice support
     "gewechat-client>=0.1.5",
     "lark-oapi>=1.5.5",
-    "mcp>=1.25.0",
+    "mcp>=1.25.0,<2.0.0",
     "nakuru-project-idk>=0.0.2.1",
     "ollama>=0.4.8",
     "openai>1.0.0",
@@ -71,7 +71,7 @@ dependencies = [
     "chromadb>=1.0.0,<2.0.0",
     "qdrant-client (>=1.15.1,<2.0.0)",
     "pyseekdb==1.1.0.post3",
-    "langbot-plugin @ git+https://github.com/langbot-app/langbot-plugin-sdk.git@1d65ed301a6afc52150a998043f73cd6032c8162",
+    "langbot-plugin @ git+https://github.com/langbot-app/langbot-plugin-sdk.git@9d216208cdfb41f0cb7fcb64632e2a46816d6dc6",
     "asyncpg>=0.30.0",
     "line-bot-sdk>=3.19.0",
     "matrix-nio>=0.25.2",

skills/skills/langbot-deploy/SKILL.md

@@ -32,12 +32,13 @@ The `all` / `box` profile starts three services:
   the LangBot and Box containers. Generate it once with `openssl rand -hex 32`;
   never put it in `box.runtime.endpoint` or commit it to config.
 
-Every Compose deployment also needs one
-`LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN` shared by `langbot` and
-`langbot_plugin_runtime`. Generate it with `openssl rand -hex 32` and export it
-before `docker compose up`; the external Plugin Runtime fails closed when the
-token is empty or weak. Kubernetes uses the `langbot-plugin-runtime-control`
-Secret shown in `docker/kubernetes.yaml`.
+A Compose deployment may optionally set
+`LANGBOT_PLUGIN_RUNTIME_CONTROL_TOKEN` on both `langbot` and
+`langbot_plugin_runtime` when port 5400 needs shared-secret protection. OSS
+defaults to leaving it unset on both sides. If enabled, generate one value with
+`openssl rand -hex 32`; configuring only one side causes the control connection
+to fail. Kubernetes may use the `langbot-plugin-runtime-control` Secret shown in
+`docker/kubernetes.yaml`.
 
 With Box off, the dashboard/skills list stays visible (read-only) but sandbox
 tools, skill add/edit, and stdio MCP are disabled. Set `box.enabled: false`

src/langbot/pkg/api/http/authz.py

@@ -19,7 +19,6 @@ class Permission(enum.StrEnum):
     WORKSPACE_VIEW = 'workspace.view'
     WORKSPACE_UPDATE = 'workspace.update'
     WORKSPACE_DELETE = 'workspace.delete'
-    OWNER_TRANSFER = 'owner.transfer'
     MEMBER_VIEW = 'member.view'
     MEMBER_INVITE = 'member.invite'
     MEMBER_UPDATE_ROLE = 'member.update_role'
@@ -49,7 +48,6 @@ _ROLE_PERMISSIONS: typing.Final = types.MappingProxyType(
             if permission
             not in {
                 Permission.WORKSPACE_DELETE,
-                Permission.OWNER_TRANSFER,
                 Permission.BILLING_LINK_MANAGE,
             }
         ),

src/langbot/pkg/api/http/controller/group.py

@@ -62,7 +62,6 @@ class AuthType(enum.Enum):
 
 _SUPPORT_ADMIN_DENIED_PERMISSIONS = frozenset(
     {
-        Permission.OWNER_TRANSFER.value,
         Permission.MEMBER_VIEW.value,
         Permission.MEMBER_INVITE.value,
         Permission.MEMBER_UPDATE_ROLE.value,

src/langbot/pkg/api/http/controller/groups/platform/bots.py

@@ -113,6 +113,24 @@ class BotsRouterGroup(group.RouterGroup):
             )
             return self.success(data={'sent': True})
 
+        @self.route(
+            '/<bot_uuid>/test-inbound',
+            methods=['POST'],
+            auth_type=group.AuthType.USER_TOKEN,
+            permission=Permission.RESOURCE_MANAGE,
+        )
+        async def _(bot_uuid: str, request_context: RequestContext) -> str:
+            json_data = await quart.request.get_json(silent=True) or {}
+            try:
+                result = await self.ap.bot_service.send_http_bot_test_message(
+                    request_context,
+                    bot_uuid,
+                    str(json_data.get('message') or ''),
+                )
+            except ValueError as exc:
+                return self.http_status(400, -1, str(exc))
+            return self.success(data=result)
+
         @self.route(
             '/<bot_uuid>/admins',
             methods=['GET'],

src/langbot/pkg/api/http/controller/groups/plugins.py

@@ -15,7 +15,6 @@ import posixpath
 import sqlalchemy
 
 from .....core import taskmgr
-from .....core.task_boundary import run_in_workspace_uow
 from .....entity.persistence import plugin as persistence_plugin
 from ...authz import Permission
 from ...context import ExecutionContext, RequestContext
@@ -311,11 +310,13 @@ class PluginsRouterGroup(group.RouterGroup):
     ):
         """Revalidate a captured task context immediately before Runtime I/O."""
 
-        await run_in_workspace_uow(
-            self.ap,
-            execution_context.workspace_uuid,
-            lambda: self.ap.plugin_connector.require_workspace_context(execution_context),
-        )
+        persistence_mgr = getattr(self.ap, 'persistence_mgr', None)
+        tenant_scope = getattr(persistence_mgr, 'tenant_scope', None)
+        if callable(tenant_scope):
+            async with tenant_scope(execution_context.workspace_uuid):
+                await self.ap.plugin_connector.require_workspace_context(execution_context)
+                return await operation()
+        await self.ap.plugin_connector.require_workspace_context(execution_context)
         return await operation()
 
     async def _require_authenticated_plugin_runtime_context(
@@ -392,8 +393,8 @@ class PluginsRouterGroup(group.RouterGroup):
         )
         async def _(request_context: RequestContext) -> str:
             """Get plugin debug information including debug URL and key"""
-            await self._require_authenticated_plugin_runtime_context(request_context)
-            debug_info = await self.ap.plugin_connector.get_debug_info()
+            execution_context = await self._require_authenticated_plugin_runtime_context(request_context)
+            debug_info = await self.ap.plugin_connector.get_debug_info(execution_context)
 
             # Get debug URL from config
             plugin_config = self.ap.instance_config.data.get('plugin', {})
@@ -403,6 +404,7 @@ class PluginsRouterGroup(group.RouterGroup):
                 data={
                     'debug_url': debug_url,
                     'plugin_debug_key': debug_info.get('plugin_debug_key', ''),
+                    'expires_at': debug_info.get('expires_at', ''),
                 }
             )
 

src/langbot/pkg/api/http/controller/groups/system.py

@@ -206,6 +206,20 @@ class SystemRouterGroup(group.RouterGroup):
 
             return self.success(data={})
 
+        @self.route(
+            '/wizard/recommended-model',
+            methods=['GET'],
+            auth_type=group.AuthType.USER_TOKEN,
+            permission=Permission.RESOURCE_MANAGE,
+        )
+        async def _(request_context: RequestContext) -> str:
+            """Resolve Space's best available chat model to this Workspace."""
+            try:
+                model = await self.ap.space_service.get_recommended_chat_model(request_context)
+            except ValueError as exc:
+                return self.http_status(503, -1, str(exc))
+            return self.success(data=model)
+
         @self.route(
             '/tasks',
             methods=['GET'],

src/langbot/pkg/api/http/controller/groups/user.py

@@ -1,6 +1,7 @@
 import quart
 import argon2
 import asyncio
+import datetime
 import uuid
 from urllib.parse import parse_qs, urlsplit
 
@@ -218,7 +219,22 @@ class UserRouterGroup(group.RouterGroup):
             try:
                 consumed_state = await self.ap.user_service.consume_space_oauth_state_details(state, 'login')
                 # Exchange code for tokens
-                token_data = await self.ap.space_service.exchange_oauth_code(code)
+                launch_workspace_uuid = consumed_state.launch_workspace_uuid
+                workspace_uuids = [launch_workspace_uuid] if launch_workspace_uuid else []
+                workspace_created_ats: dict[str, int] = {}
+                if not workspace_uuids and getattr(getattr(self.ap, 'deployment', None), 'mode', 'oss') != 'cloud':
+                    binding = await self.ap.workspace_service.get_execution_binding()
+                    workspace_uuids = [binding.workspace_uuid]
+                    workspace_created_at = binding.workspace_created_at
+                    if workspace_created_at is not None:
+                        if workspace_created_at.tzinfo is None:
+                            workspace_created_at = workspace_created_at.replace(tzinfo=datetime.UTC)
+                        workspace_created_ats[binding.workspace_uuid] = int(workspace_created_at.timestamp())
+                token_data = await self.ap.space_service.exchange_oauth_code(
+                    code,
+                    workspace_uuids,
+                    workspace_created_ats,
+                )
                 access_token = token_data.get('access_token')
                 refresh_token = token_data.get('refresh_token')
                 expires_in = token_data.get('expires_in', 0)
@@ -231,7 +247,6 @@ class UserRouterGroup(group.RouterGroup):
                     access_token, refresh_token, expires_in
                 )
 
-                launch_workspace_uuid = consumed_state.launch_workspace_uuid
                 if launch_workspace_uuid:
                     try:
                         access = await self.ap.workspace_collaboration_service.resolve_account_workspace(
@@ -285,8 +300,25 @@ class UserRouterGroup(group.RouterGroup):
                 request_context.workspace_uuid,
             )
             owner = await self.ap.user_service.get_workspace_owner(access.workspace.uuid)
-            owner_space_bound = bool(owner and owner.space_account_uuid)
-            credits = await self.ap.space_service.get_credits(owner.user) if owner_space_bound else None
+            cloud_mode = getattr(getattr(self.ap, 'deployment', None), 'mode', 'oss') == 'cloud'
+            owner_has_local_space_credentials = bool(owner and owner.space_account_uuid)
+            # Cloud Accounts authenticate through LangBot Account, so every projected
+            # Workspace owner is already bound even when this Core has no local OAuth
+            # token row (model billing uses the owner's control-plane API key).
+            owner_space_bound = cloud_mode or owner_has_local_space_credentials
+            if cloud_mode:
+                catalog_service = getattr(self.ap, 'cloud_model_catalog_service', None)
+                credits = (
+                    catalog_service.get_workspace_credits(access.workspace.uuid)
+                    if catalog_service is not None
+                    else None
+                )
+            else:
+                credits = (
+                    await self.ap.space_service.get_credits(owner.user)
+                    if owner is not None and owner.space_account_uuid
+                    else None
+                )
             return self.success(
                 data={
                     'credits': credits,
@@ -302,8 +334,10 @@ class UserRouterGroup(group.RouterGroup):
                 return self.success(data={'initialized': False})
 
             capabilities = await self.ap.user_service.get_login_capabilities()
-            if getattr(getattr(self.ap, 'deployment', None), 'mode', 'oss') == 'cloud':
+            cloud_mode = getattr(getattr(self.ap, 'deployment', None), 'mode', 'oss') == 'cloud'
+            if cloud_mode:
                 capabilities['password_login_enabled'] = False
+            capabilities['authenticated_invitation_acceptance_enabled'] = cloud_mode
             return self.success(data={'initialized': True, **capabilities})
 
         @self.route('/set-password', methods=['POST'], auth_type=group.AuthType.USER_TOKEN)

src/langbot/pkg/api/http/controller/groups/workspaces.py

@@ -30,12 +30,14 @@ def _workspace_payload(workspace: Workspace) -> dict[str, typing.Any]:
 def _membership_payload(
     membership: WorkspaceMembership,
     *,
+    display_name: str,
     email: str,
 ) -> dict[str, typing.Any]:
     return {
         'uuid': membership.uuid,
         'workspace_uuid': membership.workspace_uuid,
         'account_uuid': membership.account_uuid,
+        'display_name': display_name,
         'email': email,
         'role': membership.role,
         'status': membership.status,
@@ -94,7 +96,11 @@ class WorkspacesRouterGroup(group.RouterGroup):
                 workspaces.append(
                     {
                         'workspace': _workspace_payload(access.workspace),
-                        'membership': _membership_payload(access.membership, email=account.user),
+                        'membership': _membership_payload(
+                            access.membership,
+                            display_name=account.user,
+                            email=account.normalized_email,
+                        ),
                         'permissions': sorted(permissions_for_role(access.membership.role)),
                         'placement_generation': access.execution.placement_generation,
                         'plan_name': plan_name,
@@ -137,6 +143,7 @@ class WorkspacesRouterGroup(group.RouterGroup):
                             'uuid': None,
                             'workspace_uuid': request_context.workspace_uuid,
                             'account_uuid': None,
+                            'display_name': None,
                             'email': None,
                             'role': 'owner',
                             'status': 'active',
@@ -154,7 +161,11 @@ class WorkspacesRouterGroup(group.RouterGroup):
             return self.success(
                 data={
                     'workspace': _workspace_payload(workspace),
-                    'membership': _membership_payload(membership, email=account.user),
+                    'membership': _membership_payload(
+                        membership,
+                        display_name=account.user,
+                        email=account.normalized_email,
+                    ),
                     'permissions': sorted(request_context.workspace.permissions),
                     'placement_generation': request_context.placement_generation,
                     'plan_name': plan_name,
@@ -283,7 +294,8 @@ class WorkspacesRouterGroup(group.RouterGroup):
                 data={
                     'member': _membership_payload(
                         member,
-                        email=account.user if account is not None else '',
+                        display_name=account.user if account is not None else '',
+                        email=account.normalized_email if account is not None else '',
                     )
                 }
             )
@@ -302,7 +314,11 @@ class WorkspacesRouterGroup(group.RouterGroup):
 
     @staticmethod
     def _member_view_payload(view: WorkspaceMemberView) -> dict[str, typing.Any]:
-        return _membership_payload(view.membership, email=view.email)
+        return _membership_payload(
+            view.membership,
+            display_name=view.display_name,
+            email=view.email,
+        )
 
 
 @group.group_class('invitations', '/api/v1/invitations')

src/langbot/pkg/api/http/service/bot.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import uuid
+import json
 import sqlalchemy
 
 from ....core import app
@@ -8,6 +9,8 @@ from ....entity.persistence import bot as persistence_bot
 from ....entity.persistence import pipeline as persistence_pipeline
 from ....workspace.errors import WorkspaceNotFoundError
 from .tenant import TenantContext, require_workspace_uuid, scope_statement
+from ....utils import httpclient
+from ....platform.sources import http_bot_signing
 
 
 class BotService:
@@ -80,6 +83,7 @@ class BotService:
             'wecomcs',
             'LINE',
             'lark',
+            'http_bot',
         ]:
             webhook_prefix = self.ap.instance_config.data['api'].get('webhook_prefix', 'http://127.0.0.1:5300')
             extra_webhook_prefix = self.ap.instance_config.data['api'].get('extra_webhook_prefix', '')
@@ -216,6 +220,53 @@ class BotService:
 
         return [log.to_json() for log in logs], total_count
 
+    async def send_http_bot_test_message(
+        self,
+        context: TenantContext,
+        bot_uuid: str,
+        message: str,
+    ) -> dict:
+        """Send a signed test message through the HTTP Bot public ingress."""
+        bot = await self.get_bot(context, bot_uuid, include_secret=True)
+        if bot is None:
+            raise WorkspaceNotFoundError('Bot not found')
+        if bot.get('adapter') != 'http_bot':
+            raise ValueError('Inbound test is only available for HTTP Bot')
+        if not bot.get('enable'):
+            raise ValueError('Bot must be enabled before sending a test message')
+
+        text = message.strip()
+        if not text or len(text) > 2000:
+            raise ValueError('Test message must contain 1 to 2000 characters')
+
+        payload = {
+            'session_id': f'wizard-{uuid.uuid4().hex}',
+            'sender': {'id': 'wizard-user', 'name': 'Wizard Test'},
+            'message': [{'type': 'Plain', 'text': text}],
+        }
+        body = json.dumps(payload, ensure_ascii=False, separators=(',', ':')).encode()
+        config = bot.get('adapter_config') or {}
+        headers = {'Content-Type': 'application/json'}
+        if config.get('signature_required', True):
+            secret = str(config.get('inbound_secret') or '')
+            if not secret:
+                raise ValueError('HTTP Bot inbound signing secret is required')
+            timestamp, signature = http_bot_signing.sign(secret, body)
+            headers[http_bot_signing.HEADER_TIMESTAMP] = timestamp
+            headers[http_bot_signing.HEADER_SIGNATURE] = signature
+
+        port = int(self.ap.instance_config.data.get('api', {}).get('port', 5300))
+        session = httpclient.get_session()
+        async with session.post(
+            f'http://127.0.0.1:{port}/bots/{bot_uuid}',
+            data=body,
+            headers=headers,
+        ) as response:
+            result = await httpclient.read_json_limited(response)
+            if response.status not in {200, 202}:
+                raise ValueError(result.get('msg') or f'HTTP Bot test failed with status {response.status}')
+            return result.get('data') or {}
+
     async def send_message(
         self,
         context: TenantContext,

src/langbot/pkg/api/http/service/model.py

@@ -5,10 +5,12 @@ import uuid
 import sqlalchemy
 from langbot_plugin.api.entities.builtin.provider import message as provider_message
 
+from ....cloud.model_catalog import LANGBOT_MODELS_PROVIDER_REQUESTER
 from ....core import app
 from ....entity.persistence import model as persistence_model
 from ....entity.persistence import pipeline as persistence_pipeline
 from ....provider.modelmgr import requester as model_requester
+from ....provider.modelmgr import reasoning as model_reasoning
 from ....workspace.errors import WorkspaceNotFoundError
 from .secrets import mask_secret_value, redact_secrets, restore_secret_placeholders
 from .tenant import TenantContext, require_workspace_uuid, scope_statement
@@ -54,6 +56,53 @@ def _redact_model_secrets(model_data: dict) -> dict:
     return redacted
 
 
+def _normalize_llm_reasoning(model_data: dict) -> None:
+    model_data['reasoning_config'] = model_reasoning.validate_reasoning_config(
+        model_data.get('reasoning_config'),
+        model_data.get('abilities'),
+        model_data.get('extra_args'),
+    )
+
+
+def _validate_llm_reasoning_capability(
+    model_entity: persistence_model.LLMModel,
+    runtime_provider: model_requester.RuntimeProvider,
+) -> None:
+    config = model_reasoning.normalize_reasoning_config(model_entity.reasoning_config)
+    if config['level'] == 'provider_default':
+        return
+
+    runtime_model = model_requester.RuntimeLLMModel(
+        execution_context=runtime_provider.execution_context,
+        model_entity=model_entity,
+        provider=runtime_provider,
+    )
+    capabilities = runtime_provider.requester.get_reasoning_capabilities(runtime_model)
+    model_reasoning.validate_reasoning_capabilities(config, capabilities, model_entity.name)
+
+
+def _reasoning_capabilities(ap: app.Application, model: persistence_model.LLMModel) -> dict:
+    model_mgr = getattr(ap, 'model_mgr', None)
+    runtime_models = getattr(model_mgr, 'llm_model_dict', {}) if model_mgr is not None else {}
+    for runtime_model in runtime_models.values():
+        if (
+            runtime_model.model_entity.uuid == model.uuid
+            and runtime_model.model_entity.workspace_uuid == model.workspace_uuid
+        ):
+            return runtime_model.provider.requester.get_reasoning_capabilities(runtime_model)
+    return model_reasoning.default_reasoning_capabilities(
+        supported='reasoning' in (model.abilities or []),
+        source='manual' if 'reasoning' in (model.abilities or []) else 'unknown',
+    )
+
+
+def _serialize_llm_model(ap: app.Application, model: persistence_model.LLMModel) -> dict:
+    model_dict = ap.persistence_mgr.serialize_model(persistence_model.LLMModel, model)
+    model_dict['reasoning_config'] = model_reasoning.normalize_reasoning_config(model_dict.get('reasoning_config'))
+    model_dict['reasoning_capabilities'] = _reasoning_capabilities(ap, model)
+    return model_dict
+
+
 async def _validate_provider_supports(
     ap: app.Application,
     context: TenantContext,
@@ -113,6 +162,23 @@ async def _require_workspace_provider(
     return provider
 
 
+def _is_cloud_runtime(ap: app.Application) -> bool:
+    mode = getattr(ap.persistence_mgr, 'mode', None)
+    return getattr(mode, 'value', None) == 'cloud_runtime'
+
+
+async def _assert_cloud_managed_provider_mutable(
+    ap: app.Application,
+    context: TenantContext,
+    provider_uuid: str,
+) -> None:
+    if not _is_cloud_runtime(ap):
+        return
+    provider = await _require_workspace_provider(ap, context, provider_uuid)
+    if provider.get('requester') == LANGBOT_MODELS_PROVIDER_REQUESTER:
+        raise ValueError('LangBot Models is managed by Cloud and cannot be modified')
+
+
 async def _require_runtime_provider(
     ap: app.Application,
     context: TenantContext,
@@ -147,7 +213,7 @@ class LLMModelsService:
 
         models_list = []
         for model in models:
-            model_dict = self.ap.persistence_mgr.serialize_model(persistence_model.LLMModel, model)
+            model_dict = _serialize_llm_model(self.ap, model)
             provider = providers.get(model.provider_uuid)
             if provider:
                 provider_dict = self.ap.persistence_mgr.serialize_model(persistence_model.ModelProvider, provider)
@@ -178,7 +244,7 @@ class LLMModelsService:
             )
         )
         models = result.all()
-        serialized = [self.ap.persistence_mgr.serialize_model(persistence_model.LLMModel, m) for m in models]
+        serialized = [_serialize_llm_model(self.ap, model) for model in models]
         return serialized if include_secret else [_redact_model_secrets(model) for model in serialized]
 
     async def create_llm_model(
@@ -213,14 +279,19 @@ class LLMModelsService:
                 model_data['provider_uuid'] = provider_uuid
 
         await _require_workspace_provider(self.ap, context, model_data['provider_uuid'])
+        await _assert_cloud_managed_provider_mutable(self.ap, context, model_data['provider_uuid'])
         await _validate_provider_supports(self.ap, context, model_data['provider_uuid'], 'llm')
+        _normalize_llm_reasoning(model_data)
+
+        runtime_provider = await _require_runtime_provider(self.ap, context, model_data['provider_uuid'])
+        model_entity = persistence_model.LLMModel(**model_data)
+        _validate_llm_reasoning_capability(model_entity, runtime_provider)
 
         await self.ap.persistence_mgr.execute_async(sqlalchemy.insert(persistence_model.LLMModel).values(**model_data))
 
-        runtime_provider = await _require_runtime_provider(self.ap, context, model_data['provider_uuid'])
         runtime_llm_model = await self.ap.model_mgr.load_llm_model_with_provider(
             context,
-            persistence_model.LLMModel(**model_data),
+            model_entity,
             runtime_provider,
         )
         await self.ap.model_mgr.cache_llm_model(context, runtime_llm_model)
@@ -268,7 +339,7 @@ class LLMModelsService:
         if model is None:
             return None
 
-        model_dict = self.ap.persistence_mgr.serialize_model(persistence_model.LLMModel, model)
+        model_dict = _serialize_llm_model(self.ap, model)
 
         # Get provider
         provider_result = await self.ap.persistence_mgr.execute_async(
@@ -291,11 +362,17 @@ class LLMModelsService:
 
         return model_dict
 
-    async def update_llm_model(self, context: TenantContext, model_uuid: str, model_data: dict) -> None:
+    async def update_llm_model(
+        self,
+        context: TenantContext,
+        model_uuid: str,
+        model_data: dict,
+    ) -> None:
         """Update an existing LLM model"""
         existing_model = await self.get_llm_model(context, model_uuid, include_secret=True)
         if existing_model is None:
             raise WorkspaceNotFoundError('Model not found')
+        await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         model_data = model_data.copy()
         model_data.pop('uuid', None)
         model_data.pop('workspace_uuid', None)
@@ -321,8 +398,21 @@ class LLMModelsService:
 
         provider_uuid = model_data.get('provider_uuid', existing_model['provider_uuid'])
         await _require_workspace_provider(self.ap, context, provider_uuid)
+        await _assert_cloud_managed_provider_mutable(self.ap, context, provider_uuid)
         await _validate_provider_supports(self.ap, context, provider_uuid, 'llm')
 
+        merged_model_data = {
+            key: value
+            for key, value in {**existing_model, **model_data, 'provider_uuid': provider_uuid}.items()
+            if key not in {'provider', 'created_at', 'updated_at', 'reasoning_capabilities'}
+        }
+        _normalize_llm_reasoning(merged_model_data)
+        model_data['reasoning_config'] = merged_model_data['reasoning_config']
+
+        runtime_provider = await _require_runtime_provider(self.ap, context, provider_uuid)
+        model_entity = persistence_model.LLMModel(**_runtime_model_data(model_uuid, merged_model_data))
+        _validate_llm_reasoning_capability(model_entity, runtime_provider)
+
         result = await self.ap.persistence_mgr.execute_async(
             scope_statement(
                 sqlalchemy.update(persistence_model.LLMModel)
@@ -336,25 +426,20 @@ class LLMModelsService:
             raise WorkspaceNotFoundError('Model not found')
 
         await self.ap.model_mgr.remove_llm_model(context, model_uuid)
-        runtime_provider = await _require_runtime_provider(self.ap, context, provider_uuid)
         runtime_llm_model = await self.ap.model_mgr.load_llm_model_with_provider(
             context,
-            persistence_model.LLMModel(
-                **_runtime_model_data(
-                    model_uuid,
-                    {
-                        key: value
-                        for key, value in {**existing_model, **model_data, 'provider_uuid': provider_uuid}.items()
-                        if key not in {'provider', 'created_at', 'updated_at'}
-                    },
-                )
-            ),
+            model_entity,
             runtime_provider,
         )
         await self.ap.model_mgr.cache_llm_model(context, runtime_llm_model)
 
     async def delete_llm_model(self, context: TenantContext, model_uuid: str) -> None:
         """Delete an LLM model"""
+        if _is_cloud_runtime(self.ap):
+            existing_model = await self.get_llm_model(context, model_uuid, include_secret=True)
+            if existing_model is None:
+                raise WorkspaceNotFoundError('Model not found')
+            await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         result = await self.ap.persistence_mgr.execute_async(
             scope_statement(
                 sqlalchemy.delete(persistence_model.LLMModel).where(persistence_model.LLMModel.uuid == model_uuid),
@@ -376,6 +461,7 @@ class LLMModelsService:
                 raise WorkspaceNotFoundError('Model not found')
             runtime_llm_model = await self.ap.model_mgr.get_model_by_uuid(context, model_uuid)
         else:
+            _normalize_llm_reasoning(model_data)
             runtime_llm_model = await self.ap.model_mgr.init_temporary_runtime_llm_model(context, model_data)
 
         extra_args = model_data.get('extra_args', {})
@@ -448,7 +534,10 @@ class EmbeddingModelsService:
         return serialized if include_secret else [_redact_model_secrets(model) for model in serialized]
 
     async def create_embedding_model(
-        self, context: TenantContext, model_data: dict, preserve_uuid: bool = False
+        self,
+        context: TenantContext,
+        model_data: dict,
+        preserve_uuid: bool = False,
     ) -> str:
         """Create a new embedding model"""
         model_data = model_data.copy()
@@ -472,6 +561,7 @@ class EmbeddingModelsService:
                 model_data['provider_uuid'] = provider_uuid
 
         await _require_workspace_provider(self.ap, context, model_data['provider_uuid'])
+        await _assert_cloud_managed_provider_mutable(self.ap, context, model_data['provider_uuid'])
         await _validate_provider_supports(self.ap, context, model_data['provider_uuid'], 'text-embedding')
 
         await self.ap.persistence_mgr.execute_async(
@@ -530,11 +620,17 @@ class EmbeddingModelsService:
 
         return model_dict
 
-    async def update_embedding_model(self, context: TenantContext, model_uuid: str, model_data: dict) -> None:
+    async def update_embedding_model(
+        self,
+        context: TenantContext,
+        model_uuid: str,
+        model_data: dict,
+    ) -> None:
         """Update an existing embedding model"""
         existing_model = await self.get_embedding_model(context, model_uuid, include_secret=True)
         if existing_model is None:
             raise WorkspaceNotFoundError('Model not found')
+        await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         model_data = model_data.copy()
         model_data.pop('uuid', None)
         model_data.pop('workspace_uuid', None)
@@ -559,6 +655,7 @@ class EmbeddingModelsService:
 
         provider_uuid = model_data.get('provider_uuid', existing_model['provider_uuid'])
         await _require_workspace_provider(self.ap, context, provider_uuid)
+        await _assert_cloud_managed_provider_mutable(self.ap, context, provider_uuid)
         await _validate_provider_supports(self.ap, context, provider_uuid, 'text-embedding')
 
         result = await self.ap.persistence_mgr.execute_async(
@@ -593,6 +690,11 @@ class EmbeddingModelsService:
 
     async def delete_embedding_model(self, context: TenantContext, model_uuid: str) -> None:
         """Delete an embedding model"""
+        if _is_cloud_runtime(self.ap):
+            existing_model = await self.get_embedding_model(context, model_uuid, include_secret=True)
+            if existing_model is None:
+                raise WorkspaceNotFoundError('Model not found')
+            await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         result = await self.ap.persistence_mgr.execute_async(
             scope_statement(
                 sqlalchemy.delete(persistence_model.EmbeddingModel).where(
@@ -685,7 +787,12 @@ class RerankModelsService:
         serialized = [self.ap.persistence_mgr.serialize_model(persistence_model.RerankModel, m) for m in models]
         return serialized if include_secret else [_redact_model_secrets(model) for model in serialized]
 
-    async def create_rerank_model(self, context: TenantContext, model_data: dict, preserve_uuid: bool = False) -> str:
+    async def create_rerank_model(
+        self,
+        context: TenantContext,
+        model_data: dict,
+        preserve_uuid: bool = False,
+    ) -> str:
         """Create a new rerank model"""
         model_data = model_data.copy()
         if not preserve_uuid:
@@ -708,6 +815,7 @@ class RerankModelsService:
                 model_data['provider_uuid'] = provider_uuid
 
         await _require_workspace_provider(self.ap, context, model_data['provider_uuid'])
+        await _assert_cloud_managed_provider_mutable(self.ap, context, model_data['provider_uuid'])
         await _validate_provider_supports(self.ap, context, model_data['provider_uuid'], 'rerank')
 
         await self.ap.persistence_mgr.execute_async(
@@ -766,11 +874,17 @@ class RerankModelsService:
 
         return model_dict
 
-    async def update_rerank_model(self, context: TenantContext, model_uuid: str, model_data: dict) -> None:
+    async def update_rerank_model(
+        self,
+        context: TenantContext,
+        model_uuid: str,
+        model_data: dict,
+    ) -> None:
         """Update an existing rerank model"""
         existing_model = await self.get_rerank_model(context, model_uuid, include_secret=True)
         if existing_model is None:
             raise WorkspaceNotFoundError('Model not found')
+        await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         model_data = model_data.copy()
         model_data.pop('uuid', None)
         model_data.pop('workspace_uuid', None)
@@ -795,6 +909,7 @@ class RerankModelsService:
 
         provider_uuid = model_data.get('provider_uuid', existing_model['provider_uuid'])
         await _require_workspace_provider(self.ap, context, provider_uuid)
+        await _assert_cloud_managed_provider_mutable(self.ap, context, provider_uuid)
         await _validate_provider_supports(self.ap, context, provider_uuid, 'rerank')
 
         result = await self.ap.persistence_mgr.execute_async(
@@ -829,6 +944,11 @@ class RerankModelsService:
 
     async def delete_rerank_model(self, context: TenantContext, model_uuid: str) -> None:
         """Delete a rerank model"""
+        if _is_cloud_runtime(self.ap):
+            existing_model = await self.get_rerank_model(context, model_uuid, include_secret=True)
+            if existing_model is None:
+                raise WorkspaceNotFoundError('Model not found')
+            await _assert_cloud_managed_provider_mutable(self.ap, context, existing_model['provider_uuid'])
         result = await self.ap.persistence_mgr.execute_async(
             scope_statement(
                 sqlalchemy.delete(persistence_model.RerankModel).where(

src/langbot/pkg/api/http/service/provider.py

@@ -5,6 +5,7 @@ import traceback
 
 import sqlalchemy
 
+from ....cloud.model_catalog import LANGBOT_MODELS_PROVIDER_REQUESTER
 from ....core import app
 from ....entity.persistence import model as persistence_model
 from ....workspace.errors import WorkspaceNotFoundError
@@ -20,6 +21,20 @@ class ModelProviderService:
     def __init__(self, ap: app.Application) -> None:
         self.ap = ap
 
+    def _is_cloud_runtime(self) -> bool:
+        mode = getattr(self.ap.persistence_mgr, 'mode', None)
+        return getattr(mode, 'value', None) == 'cloud_runtime'
+
+    def _system_requester_is_reserved(self, requester: object) -> bool:
+        return self._is_cloud_runtime() and requester == LANGBOT_MODELS_PROVIDER_REQUESTER
+
+    async def _assert_provider_mutable(self, context: TenantContext, provider_uuid: str) -> None:
+        if not self._is_cloud_runtime():
+            return
+        provider = await self.get_provider(context, provider_uuid)
+        if provider is not None and self._system_requester_is_reserved(provider.get('requester')):
+            raise ValueError('LangBot Models is managed by Cloud and cannot be modified')
+
     @staticmethod
     def _normalize_api_keys(api_keys: str | list[str] | tuple[str, ...] | None) -> list[str]:
         if api_keys is None:
@@ -99,6 +114,8 @@ class ModelProviderService:
     async def create_provider(self, context: TenantContext, provider_data: dict) -> str:
         """Create a new provider"""
         provider_data = provider_data.copy()
+        if self._system_requester_is_reserved(provider_data.get('requester')):
+            raise ValueError('space-chat-completions is reserved for the Cloud-managed LangBot Models provider')
         provider_data['uuid'] = str(uuid.uuid4())
         provider_data['workspace_uuid'] = require_workspace_uuid(context)
         provider_data['api_keys'] = self._normalize_api_keys(
@@ -115,7 +132,10 @@ class ModelProviderService:
 
     async def update_provider(self, context: TenantContext, provider_uuid: str, provider_data: dict) -> None:
         """Update an existing provider"""
+        await self._assert_provider_mutable(context, provider_uuid)
         provider_data = provider_data.copy()
+        if self._system_requester_is_reserved(provider_data.get('requester')):
+            raise ValueError('space-chat-completions is reserved for the Cloud-managed LangBot Models provider')
         provider_data.pop('uuid', None)
         provider_data.pop('workspace_uuid', None)
         if 'api_keys' in provider_data:
@@ -145,6 +165,7 @@ class ModelProviderService:
 
     async def delete_provider(self, context: TenantContext, provider_uuid: str) -> None:
         """Delete a provider (only if no models reference it)"""
+        await self._assert_provider_mutable(context, provider_uuid)
         workspace_uuid = require_workspace_uuid(context)
         # Check if any models use this provider
         llm_result = await self.ap.persistence_mgr.execute_async(
@@ -245,6 +266,8 @@ class ModelProviderService:
         api_keys: list,
     ) -> str:
         """Find existing provider or create new one"""
+        if self._system_requester_is_reserved(requester):
+            raise ValueError('space-chat-completions is reserved for the Cloud-managed LangBot Models provider')
         workspace_uuid = require_workspace_uuid(context)
         api_keys = self._normalize_api_keys(restore_secret_placeholders(api_keys, sensitive=True))
 

src/langbot/pkg/api/http/service/space.py

@@ -11,6 +11,9 @@ import sqlalchemy
 from ....core import app
 from ....entity.persistence import user
 from ....entity.dto.space_model import SpaceModel
+from ....entity.dto.space_model import SpaceModelSelection
+from ....entity.persistence import model as persistence_model
+from ....cloud.model_catalog import LANGBOT_MODELS_PROVIDER_REQUESTER
 
 
 _CREDITS_CACHE_TTL_SECONDS = 60
@@ -59,6 +62,10 @@ class SpaceService:
         result_list = result.all()
         return result_list[0] if result_list else None
 
+    async def get_valid_access_token(self, user_email: str) -> str | None:
+        """Return a current Space bearer, refreshing and persisting it when needed."""
+        return await self._ensure_valid_token(user_email)
+
     async def _ensure_valid_token(self, user_email: str) -> str | None:
         """Ensure access token is valid, refresh if expired. Returns valid access_token or None."""
         user_obj = await self._get_user_by_email(user_email)
@@ -117,7 +124,12 @@ class SpaceService:
             params['state'] = state
         return f'{authorize_url}?{urlencode(params)}'
 
-    async def exchange_oauth_code(self, code: str) -> typing.Dict:
+    async def exchange_oauth_code(
+        self,
+        code: str,
+        workspace_uuids: list[str] | None = None,
+        workspace_created_ats: dict[str, int] | None = None,
+    ) -> typing.Dict:
         """Exchange OAuth authorization code for tokens"""
         from langbot.pkg.utils import constants
 
@@ -127,7 +139,14 @@ class SpaceService:
         session = httpclient.get_session()
         async with session.post(
             f'{space_url}/api/v1/accounts/oauth/token',
-            json={'code': code, 'instance_id': constants.instance_id},
+            json={
+                'code': code,
+                'instance_id': constants.instance_id,
+                # Sending an explicit empty list tells new Space servers not to
+                # synthesize a legacy instance-derived Workspace binding.
+                'workspace_uuids': workspace_uuids if workspace_uuids is not None else [],
+                'workspace_created_ats': workspace_created_ats or {},
+            },
         ) as response:
             if response.status != 200:
                 error = await httpclient.read_text_limited(response)
@@ -222,3 +241,79 @@ class SpaceService:
                 raise ValueError(f'Failed to get models: {data.get("msg")}')
             models_data = data.get('data', {}).get('models', [])
             return [SpaceModel.model_validate(model_dict) for model_dict in models_data]
+
+    async def get_model_selection(self, category: str) -> typing.List[SpaceModelSelection]:
+        """Return Space models in the availability-ranked selection order."""
+        space_url = self._get_space_config()['url']
+        session = httpclient.get_session()
+        async with session.get(
+            f'{space_url}/api/v1/models/selection',
+            params={'category': category},
+        ) as response:
+            if response.status != 200:
+                error = await httpclient.read_text_limited(response)
+                raise ValueError(f'Failed to get model selection: {error}')
+            payload = await httpclient.read_json_limited(response)
+            if payload.get('code') != 0:
+                raise ValueError(f'Failed to get model selection: {payload.get("msg")}')
+
+            data = payload.get('data', [])
+            if isinstance(data, dict):
+                data = data.get('models', data.get('items', []))
+            if not isinstance(data, list):
+                raise ValueError('Failed to get model selection: invalid response')
+
+            models = []
+            for selection in data:
+                if isinstance(selection, dict) and isinstance(selection.get('model'), dict):
+                    models.append(selection['model'])
+                else:
+                    models.append(selection)
+            return [SpaceModelSelection.model_validate(model) for model in models]
+
+    async def get_recommended_chat_model(self, context: typing.Any) -> dict:
+        """Resolve Space's first ranked chat model to a local Workspace model."""
+        selection = await self.get_model_selection('chat')
+        if not selection:
+            raise ValueError('No recommended chat model is available')
+        recommended = selection[0]
+
+        async def find_local_model():
+            result = await self.ap.persistence_mgr.execute_async(
+                sqlalchemy.select(persistence_model.LLMModel)
+                .join(
+                    persistence_model.ModelProvider,
+                    sqlalchemy.and_(
+                        persistence_model.ModelProvider.workspace_uuid
+                        == persistence_model.LLMModel.workspace_uuid,
+                        persistence_model.ModelProvider.uuid == persistence_model.LLMModel.provider_uuid,
+                    ),
+                )
+                .where(
+                    persistence_model.LLMModel.workspace_uuid == context.workspace_uuid,
+                    persistence_model.ModelProvider.requester == LANGBOT_MODELS_PROVIDER_REQUESTER,
+                    sqlalchemy.or_(
+                        persistence_model.LLMModel.uuid == recommended.uuid,
+                        persistence_model.LLMModel.name == recommended.model_id,
+                    ),
+                )
+            )
+            return result.first()
+
+        local_model = await find_local_model()
+        if local_model is None:
+            # OSS synchronizes the public catalog locally. Refresh once in case
+            # the recommendation was published after this process started.
+            from ..context import ExecutionContext
+
+            try:
+                await self.ap.model_mgr.sync_new_models_from_space(
+                    ExecutionContext.from_request(context)
+                )
+            except Exception:
+                pass
+            local_model = await find_local_model()
+
+        if local_model is None:
+            raise ValueError('Recommended chat model is not available in this Workspace')
+        return {'uuid': local_model.uuid, 'name': local_model.name}

src/langbot/pkg/api/http/service/user.py

@@ -779,8 +779,27 @@ class UserService:
         local_account = await self.get_user_by_email(user_email)
         if local_account is None:
             raise ValueError('User not found')
-        # Exchange code for tokens
-        token_data = await self.ap.space_service.exchange_oauth_code(code)
+        # Exchange code for tokens and bind both installation and the active
+        # OSS Workspace as independent identities.
+        workspace_service = getattr(self.ap, 'workspace_service', None)
+        if workspace_service is not None:
+            binding = await workspace_service.get_execution_binding()
+            created_at = binding.workspace_created_at
+            created_ts = (
+                int(created_at.replace(tzinfo=datetime.timezone.utc).timestamp())
+                if created_at.tzinfo is None
+                else int(created_at.timestamp())
+            )
+            token_data = await self.ap.space_service.exchange_oauth_code(
+                code,
+                [binding.workspace_uuid],
+                {binding.workspace_uuid: created_ts},
+            )
+        else:
+            # Compatibility for early/bootstrap call sites that have not wired
+            # WorkspaceService yet; old Space servers still derive the legacy
+            # Workspace identity from instance_id when the field is omitted.
+            token_data = await self.ap.space_service.exchange_oauth_code(code)
         access_token = token_data.get('access_token')
         refresh_token = token_data.get('refresh_token')
         expires_in = token_data.get('expires_in', 0)

src/langbot/pkg/cloud/bootstrap.py

@@ -13,11 +13,12 @@ from typing import Any, Protocol, runtime_checkable
 from ..workspace.policy import CloudWorkspacePolicy, SingleWorkspacePolicy
 from .directory import DirectoryProjectionProvider, directory_projection_limits_from_config
 from .entitlements import EntitlementProvider, OpenSourceEntitlementProvider
+from .model_catalog import CloudModelCatalogProvider
 
 
 CLOUD_BOOTSTRAP_ENTRY_POINT = 'langbot.cloud_bootstrap'
 REQUIRED_TENANT_ISOLATION_VERSION = 2
-SUPPORTED_PGVECTOR_DIMENSIONS = frozenset({384, 512, 768, 1024, 1536})
+SUPPORTED_PGVECTOR_DIMENSIONS = frozenset({384, 512, 768, 1024, 1536, 3072})
 
 
 class CloudBootstrapError(RuntimeError):
@@ -50,6 +51,7 @@ class OpenSourceDeployment:
     )
     directory_provider: None = None
     manifest_provider: None = None
+    model_catalog_provider: None = None
     persistence_mode: str = 'oss_compat'
     required_vector_backend: str | None = None
 
@@ -80,6 +82,7 @@ class VerifiedCloudDeployment:
     entitlement_provider: EntitlementProvider
     directory_provider: DirectoryProjectionProvider
     manifest_provider: CloudManifestProvider
+    model_catalog_provider: CloudModelCatalogProvider
     verification_key_id: str
     mode: str = dataclasses.field(default='cloud', init=False)
     workspace_policy: CloudWorkspacePolicy = dataclasses.field(default_factory=CloudWorkspacePolicy, init=False)
@@ -110,6 +113,8 @@ class VerifiedCloudDeployment:
             raise CloudBootstrapError('Verified Cloud bootstrap did not provide a directory adapter')
         if not isinstance(self.manifest_provider, CloudManifestProvider):
             raise CloudBootstrapError('Verified Cloud bootstrap did not provide a Manifest renewal adapter')
+        if not isinstance(self.model_catalog_provider, CloudModelCatalogProvider):
+            raise CloudBootstrapError('Verified Cloud bootstrap did not provide a model catalog adapter')
 
     def validate_instance_config(self, config: dict[str, Any]) -> None:
         try:
@@ -138,8 +143,14 @@ class VerifiedCloudDeployment:
         if plugin_worker.get('require_hard_limits') is not True:
             raise CloudBootstrapError('Cloud Runtime requires plugin.worker.require_hard_limits=true')
         box_config = config.get('box', {})
-        if box_config.get('enabled') is not True:
-            raise CloudBootstrapError('Cloud runtime requires box.enabled=true')
+        box_enabled = box_config.get('enabled')
+        if box_enabled is False:
+            # Explicitly disabling Box removes the sandbox surface entirely and
+            # therefore does not weaken tenant isolation. Validate the strict
+            # runtime/admission contract only when the surface is enabled.
+            return
+        if box_enabled is not True:
+            raise CloudBootstrapError('Cloud runtime requires box.enabled to be an explicit boolean')
         if box_config.get('backend') != 'nsjail':
             raise CloudBootstrapError('Cloud runtime requires box.backend=nsjail')
         runtime_endpoint = str(box_config.get('runtime', {}).get('endpoint', '') or '').strip()

src/langbot/pkg/cloud/directory_projection.py

@@ -15,6 +15,7 @@ from ..entity.persistence.cloud_directory import DirectoryProjectionInbox, Direc
 from ..entity.persistence.user import AccountSource, AccountStatus, User
 from ..entity.persistence.workspace import (
     MembershipRole,
+    MembershipSource,
     MembershipStatus,
     Workspace,
     WorkspaceExecutionSource,
@@ -358,6 +359,7 @@ class DirectoryProjectionService:
 
         await self._reconcile_entitlement_snapshot_set(snapshot)
         self._publish_runtime_execution_projection(snapshot.workspaces)
+        self._request_model_catalog_sync()
         self._record_batch_cardinality(
             active_workspaces=active_workspace_count,
             workspaces=workspace_count,
@@ -466,6 +468,7 @@ class DirectoryProjectionService:
             returned.values(),
             affected_workspace_uuids=requested,
         )
+        self._request_model_catalog_sync()
         self._record_batch_cardinality(
             active_workspaces=active_workspace_count,
             workspaces=workspace_count,
@@ -475,6 +478,14 @@ class DirectoryProjectionService:
             self._record_success()
         self._consumer_cursor = batch.cursor
 
+    def _request_model_catalog_sync(self) -> None:
+        """Wake model provisioning after a committed directory change."""
+
+        service = getattr(self.ap, 'cloud_model_catalog_service', None)
+        request_sync = getattr(service, 'request_sync', None)
+        if callable(request_sync):
+            request_sync()
+
     def _publish_runtime_execution_projection(
         self,
         workspaces: Iterable[DirectoryWorkspace],
@@ -876,15 +887,15 @@ class DirectoryProjectionService:
                         account_uuid=member.account_uuid,
                         role=role,
                         status=status,
+                        source=MembershipSource.CLOUD_PROJECTION.value,
                         joined_at=joined_at,
                         projection_revision=member.projection_revision,
                     )
                 )
                 continue
-            if membership.projection_revision == 0:
-                # Revision zero is Core-owned collaboration state. Directory
-                # projection seeds memberships, but must not overwrite later
-                # invitation, role, or removal decisions made by Core.
+            if membership.source != MembershipSource.CLOUD_PROJECTION.value:
+                # Core-owned collaboration state is never adopted based on
+                # account provenance, revision, or matching account identity.
                 continue
             if membership.uuid != member.membership_uuid:
                 raise DirectoryProjectionUnavailableError('Directory membership UUID changed for one account')
@@ -896,11 +907,12 @@ class DirectoryProjectionService:
                 raise DirectoryProjectionUnavailableError('Directory membership revision has conflicting contents')
             membership.role = role
             membership.status = status
+            membership.source = MembershipSource.CLOUD_PROJECTION.value
             membership.joined_at = joined_at
             membership.projection_revision = member.projection_revision
 
         for account_uuid, membership in existing.items():
-            if account_uuid not in included_accounts and membership.projection_revision != 0:
+            if account_uuid not in included_accounts and membership.source == MembershipSource.CLOUD_PROJECTION.value:
                 membership.status = MembershipStatus.REMOVED.value
                 membership.projection_revision = max(
                     int(membership.projection_revision),

src/langbot/pkg/cloud/model_catalog.py

@@ -0,0 +1,337 @@
+from __future__ import annotations
+
+import asyncio
+import uuid
+from datetime import datetime
+from typing import Any, Literal, Protocol, runtime_checkable
+
+import sqlalchemy
+from pydantic import BaseModel, ConfigDict, Field, SecretStr, field_validator
+
+from ..entity.persistence import model as persistence_model
+
+
+LANGBOT_MODELS_PROVIDER_REQUESTER = 'space-chat-completions'
+LANGBOT_MODELS_PROVIDER_NAME = 'LangBot Models'
+_MODEL_RESOURCE_NAMESPACE = uuid.UUID('94c703ca-1df5-4e91-bcd3-74ac65cb7921')
+_SUPPORTED_CATEGORIES = {'chat', 'embedding', 'rerank'}
+_MODEL_TABLES = (
+    persistence_model.LLMModel,
+    persistence_model.EmbeddingModel,
+    persistence_model.RerankModel,
+)
+
+
+class CloudModelCatalogItem(BaseModel):
+    model_config = ConfigDict(extra='forbid', frozen=True)
+
+    uuid: str = Field(min_length=1, max_length=255)
+    model_id: str = Field(min_length=1, max_length=255)
+    category: Literal['chat', 'embedding', 'rerank']
+    llm_abilities: tuple[str, ...] = ()
+    is_featured: bool = False
+    featured_order: int = 0
+
+    @field_validator('llm_abilities', mode='before')
+    @classmethod
+    def normalize_missing_abilities(cls, value: Any) -> Any:
+        return () if value is None else value
+
+    @field_validator('llm_abilities')
+    @classmethod
+    def validate_abilities(cls, value: tuple[str, ...]) -> tuple[str, ...]:
+        if any(not item.strip() or len(item) > 64 for item in value):
+            raise ValueError('Model abilities must be non-empty strings of at most 64 characters')
+        if len(set(value)) != len(value):
+            raise ValueError('Model abilities must be unique')
+        return value
+
+
+class CloudWorkspaceModelBilling(BaseModel):
+    model_config = ConfigDict(extra='forbid', frozen=True)
+
+    workspace_uuid: str = Field(min_length=36, max_length=36)
+    owner_account_uuid: str | None = Field(default=None, min_length=36, max_length=36)
+    api_key: SecretStr | None = None
+    credits: int | None = None
+
+    @field_validator('workspace_uuid')
+    @classmethod
+    def validate_uuid(cls, value: str) -> str:
+        return str(uuid.UUID(value))
+
+    @field_validator('owner_account_uuid')
+    @classmethod
+    def validate_optional_uuid(cls, value: str | None) -> str | None:
+        return None if value is None else str(uuid.UUID(value))
+
+
+class CloudModelCatalogSnapshot(BaseModel):
+    model_config = ConfigDict(extra='forbid', frozen=True)
+
+    instance_uuid: str = Field(min_length=1, max_length=255)
+    generated_at: datetime
+    base_url: str = Field(min_length=1, max_length=512)
+    models: tuple[CloudModelCatalogItem, ...]
+    workspaces: tuple[CloudWorkspaceModelBilling, ...]
+
+    @field_validator('base_url')
+    @classmethod
+    def validate_base_url(cls, value: str) -> str:
+        normalized = value.rstrip('/')
+        if not normalized.startswith('https://'):
+            raise ValueError('Cloud model gateway base URL must use HTTPS')
+        return normalized
+
+    @field_validator('models')
+    @classmethod
+    def validate_models(cls, value: tuple[CloudModelCatalogItem, ...]) -> tuple[CloudModelCatalogItem, ...]:
+        if len(value) > 500:
+            raise ValueError('Cloud model catalog exceeds 500 models')
+        identities = {(item.category, item.uuid) for item in value}
+        if len(identities) != len(value):
+            raise ValueError('Cloud model catalog contains duplicate model identities')
+        return value
+
+    @field_validator('workspaces')
+    @classmethod
+    def validate_workspaces(
+        cls, value: tuple[CloudWorkspaceModelBilling, ...]
+    ) -> tuple[CloudWorkspaceModelBilling, ...]:
+        if len(value) > 10_000:
+            raise ValueError('Cloud model catalog exceeds 10000 Workspaces')
+        identities = {item.workspace_uuid for item in value}
+        if len(identities) != len(value):
+            raise ValueError('Cloud model catalog contains duplicate Workspaces')
+        return value
+
+
+@runtime_checkable
+class CloudModelCatalogProvider(Protocol):
+    async def fetch_model_catalog(self, instance_uuid: str) -> CloudModelCatalogSnapshot:
+        """Fetch and verify the complete model catalog and Workspace billing projection."""
+        ...
+
+
+def system_provider_uuid(workspace_uuid: str) -> str:
+    workspace = str(uuid.UUID(workspace_uuid))
+    return str(uuid.uuid5(_MODEL_RESOURCE_NAMESPACE, f'{workspace}:provider:{LANGBOT_MODELS_PROVIDER_REQUESTER}'))
+
+
+def system_model_uuid(workspace_uuid: str, category: str, upstream_uuid: str) -> str:
+    workspace = str(uuid.UUID(workspace_uuid))
+    if category not in _SUPPORTED_CATEGORIES:
+        raise ValueError(f'Unsupported model category: {category}')
+    if not upstream_uuid:
+        raise ValueError('Upstream model UUID is required')
+    return str(uuid.uuid5(_MODEL_RESOURCE_NAMESPACE, f'{workspace}:model:{category}:{upstream_uuid}'))
+
+
+class CloudModelCatalogSyncService:
+    """Reconcile Space-owned model catalog and Owner billing tokens into every Cloud Workspace."""
+
+    def __init__(
+        self,
+        ap: Any,
+        provider: CloudModelCatalogProvider,
+        instance_uuid: str,
+        *,
+        sync_interval_seconds: float = 3600.0,
+    ) -> None:
+        if not isinstance(provider, CloudModelCatalogProvider):
+            raise TypeError('Cloud model catalog sync requires a CloudModelCatalogProvider')
+        if sync_interval_seconds < 10:
+            raise ValueError('Cloud model catalog sync interval must be at least 10 seconds')
+        self.ap = ap
+        self.provider = provider
+        self.instance_uuid = instance_uuid
+        self.sync_interval_seconds = float(sync_interval_seconds)
+        # A tenant UoW commits one Workspace at a time. Keep a durable in-memory
+        # convergence marker so a failed runtime reload is retried even when the
+        # following database reconciliation is a no-op.
+        self._runtime_reload_pending = False
+        self._workspace_credits: dict[str, int | None] = {}
+        self._sync_requested = asyncio.Event()
+
+    def get_workspace_credits(self, workspace_uuid: str) -> int | None:
+        """Return the latest signed owner-credit projection for a Workspace."""
+        return self._workspace_credits.get(str(uuid.UUID(workspace_uuid)))
+
+    async def initialize(self) -> None:
+        await self.sync_once(reload_runtime=False)
+
+    def request_sync(self) -> None:
+        """Wake the catalog loop after a directory Workspace change."""
+
+        self._sync_requested.set()
+
+    async def run(self) -> None:
+        while True:
+            try:
+                await asyncio.wait_for(self._sync_requested.wait(), timeout=self.sync_interval_seconds)
+            except TimeoutError:
+                pass
+            self._sync_requested.clear()
+            try:
+                await self.sync_once(reload_runtime=True)
+            except asyncio.CancelledError:
+                raise
+            except Exception as exc:
+                # Exception messages can contain rendered SQL bound values,
+                # including provider API keys. Log only the exception class.
+                self.ap.logger.warning(f'Cloud model catalog synchronization failed ({type(exc).__name__})')
+
+    async def sync_once(self, *, reload_runtime: bool = True) -> dict[str, int]:
+        summary = {'workspaces': 0, 'created': 0, 'updated': 0, 'deleted': 0}
+        snapshot: CloudModelCatalogSnapshot | None = None
+        sync_error: Exception | None = None
+        reload_error: Exception | None = None
+        try:
+            snapshot = await self.provider.fetch_model_catalog(self.instance_uuid)
+            if snapshot.instance_uuid != self.instance_uuid:
+                raise ValueError('Cloud model catalog targets another LangBot instance')
+
+            bindings = await self.ap.workspace_service.list_active_execution_bindings()
+            billing_by_workspace = {item.workspace_uuid: item for item in snapshot.workspaces}
+            missing = sorted(
+                binding.workspace_uuid for binding in bindings if binding.workspace_uuid not in billing_by_workspace
+            )
+            if missing:
+                raise ValueError(
+                    f'Cloud model catalog is missing billing projections for {len(missing)} active Workspaces'
+                )
+
+            for binding in bindings:
+                counts = await self._sync_workspace(
+                    binding.workspace_uuid,
+                    snapshot,
+                    billing_by_workspace[binding.workspace_uuid],
+                )
+                summary['workspaces'] += 1
+                workspace_changed = any(counts[key] > 0 for key in ('created', 'updated', 'deleted'))
+                if workspace_changed:
+                    # _sync_workspace returns only after its tenant UoW commits.
+                    self._runtime_reload_pending = True
+                for key in ('created', 'updated', 'deleted'):
+                    summary[key] += counts[key]
+                self._workspace_credits[binding.workspace_uuid] = billing_by_workspace[binding.workspace_uuid].credits
+        except Exception as exc:
+            sync_error = exc
+        finally:
+            model_mgr = getattr(self.ap, 'model_mgr', None)
+            if reload_runtime and self._runtime_reload_pending and model_mgr is not None:
+                try:
+                    await model_mgr.load_models_from_db()
+                except Exception as exc:
+                    reload_error = exc
+                else:
+                    self._runtime_reload_pending = False
+
+        if sync_error is not None:
+            if reload_error is not None:
+                raise sync_error from reload_error
+            raise sync_error
+        if reload_error is not None:
+            raise reload_error
+
+        changed = any(summary[key] > 0 for key in ('created', 'updated', 'deleted'))
+        if changed and snapshot is not None:
+            self.ap.logger.info(
+                'Cloud model catalog synchronized '
+                f'({summary["workspaces"]} Workspaces, {len(snapshot.models)} models, '
+                f'created={summary["created"]}, updated={summary["updated"]}, deleted={summary["deleted"]})'
+            )
+        return summary
+
+    async def _sync_workspace(
+        self,
+        workspace_uuid: str,
+        snapshot: CloudModelCatalogSnapshot,
+        billing: CloudWorkspaceModelBilling,
+    ) -> dict[str, int]:
+        counts = {'created': 0, 'updated': 0, 'deleted': 0}
+        provider_uuid = system_provider_uuid(workspace_uuid)
+        desired_keys = [billing.api_key.get_secret_value()] if billing.api_key is not None else []
+
+        async with self.ap.persistence_mgr.tenant_uow(workspace_uuid) as uow:
+            provider = await uow.session.scalar(
+                sqlalchemy.select(persistence_model.ModelProvider).where(
+                    persistence_model.ModelProvider.uuid == provider_uuid
+                )
+            )
+            provider_values = {
+                'workspace_uuid': workspace_uuid,
+                'name': LANGBOT_MODELS_PROVIDER_NAME,
+                'requester': LANGBOT_MODELS_PROVIDER_REQUESTER,
+                'base_url': snapshot.base_url,
+                'api_keys': desired_keys,
+            }
+            if provider is None:
+                provider = persistence_model.ModelProvider(uuid=provider_uuid, **provider_values)
+                uow.session.add(provider)
+                await uow.session.flush()
+                counts['created'] += 1
+            elif self._update_entity(provider, provider_values):
+                counts['updated'] += 1
+
+            existing_by_table: dict[type, dict[str, Any]] = {}
+            for table in _MODEL_TABLES:
+                rows = (
+                    await uow.session.scalars(sqlalchemy.select(table).where(table.provider_uuid == provider_uuid))
+                ).all()
+                existing_by_table[table] = {row.uuid: row for row in rows}
+
+            desired_ids: dict[type, set[str]] = {table: set() for table in _MODEL_TABLES}
+            for item in snapshot.models:
+                table, values = self._model_values(workspace_uuid, provider_uuid, item)
+                model_uuid = system_model_uuid(workspace_uuid, item.category, item.uuid)
+                desired_ids[table].add(model_uuid)
+                existing = existing_by_table[table].get(model_uuid)
+                if existing is None:
+                    uow.session.add(table(uuid=model_uuid, **values))
+                    counts['created'] += 1
+                elif self._update_entity(existing, values):
+                    counts['updated'] += 1
+
+            for table, entities in existing_by_table.items():
+                for model_uuid, entity in entities.items():
+                    if model_uuid not in desired_ids[table]:
+                        await uow.session.delete(entity)
+                        counts['deleted'] += 1
+
+        return counts
+
+    @staticmethod
+    def _update_entity(entity: Any, values: dict[str, Any]) -> bool:
+        changed = False
+        for key, value in values.items():
+            if getattr(entity, key) != value:
+                setattr(entity, key, value)
+                changed = True
+        return changed
+
+    @staticmethod
+    def _model_values(
+        workspace_uuid: str,
+        provider_uuid: str,
+        item: CloudModelCatalogItem,
+    ) -> tuple[type, dict[str, Any]]:
+        ranking = 100 - item.featured_order if item.is_featured else 0
+        common = {
+            'workspace_uuid': workspace_uuid,
+            'name': item.model_id,
+            'provider_uuid': provider_uuid,
+            'extra_args': {},
+            'prefered_ranking': ranking,
+        }
+        if item.category == 'chat':
+            return persistence_model.LLMModel, {
+                **common,
+                'abilities': list(item.llm_abilities),
+                'context_length': None,
+            }
+        if item.category == 'embedding':
+            return persistence_model.EmbeddingModel, common
+        if item.category == 'rerank':
+            return persistence_model.RerankModel, common
+        raise ValueError(f'Unsupported model category: {item.category}')

src/langbot/pkg/core/app.py

@@ -54,6 +54,7 @@ from ..cloud import launch as cloud_launch_module
 from ..cloud import support_admin as cloud_support_admin_module
 from ..cloud import directory_projection as cloud_directory_projection_module
 from ..cloud import entitlements as cloud_entitlements_module
+from ..cloud import model_catalog as cloud_model_catalog_module
 from ..api.http.context import ExecutionContext, PrincipalContext, PrincipalType
 
 
@@ -142,13 +143,12 @@ class Application:
     deployment: cloud_bootstrap_module.OpenSourceDeployment | cloud_bootstrap_module.VerifiedCloudDeployment = None
 
     deployment_admission: cloud_bootstrap_module.DeploymentAdmissionGuard = None
-
+    directory_projection_service: cloud_directory_projection_module.DirectoryProjectionService | None = None
+    cloud_model_catalog_service: cloud_model_catalog_module.CloudModelCatalogSyncService | None = None
     manifest_refresh_service: cloud_bootstrap_module.CloudManifestRefreshService | None = None
 
     entitlement_resolver: cloud_entitlements_module.EntitlementResolver | None = None
 
-    directory_projection_service: cloud_directory_projection_module.DirectoryProjectionService | None = None
-
     vector_db_mgr: vectordb_mgr.VectorDBManager = None
 
     http_ctrl: http_controller.HTTPController = None
@@ -306,6 +306,12 @@ class Application:
                     name='cloud-directory-projection',
                     scopes=[core_entities.LifecycleControlScope.APPLICATION],
                 )
+            if self.cloud_model_catalog_service is not None:
+                self.task_mgr.create_task(
+                    self.cloud_model_catalog_service.run(),
+                    name='cloud-model-catalog-sync',
+                    scopes=[core_entities.LifecycleControlScope.APPLICATION],
+                )
             if self.manifest_refresh_service is not None:
                 self.task_mgr.create_task(
                     self.manifest_refresh_service.run(),

src/langbot/pkg/core/stages/build_app.py

@@ -46,6 +46,7 @@ from ...cloud import support_admin as cloud_support_admin_module
 from ...cloud.directory import directory_projection_limits_from_config
 from ...cloud.directory_projection import DirectoryProjectionService
 from ...cloud.entitlements import EntitlementResolver
+from ...cloud.model_catalog import CloudModelCatalogSyncService
 from ...api.http.context import ExecutionContext, PrincipalContext, PrincipalType
 from ...api.http.authz import WorkspaceRequiredError
 
@@ -176,6 +177,16 @@ class BuildAppStage(stage.BootingStage):
             # of repeating tenant validation for every manager.
             await workspace_service_inst.prime_startup_execution_bindings()
 
+            if not isinstance(deployment, cloud_bootstrap.VerifiedCloudDeployment):
+                raise RuntimeError('Multi-Workspace runtime requires a verified Cloud deployment')
+            cloud_model_catalog_service = CloudModelCatalogSyncService(
+                ap,
+                deployment.model_catalog_provider,
+                constants.instance_id,
+            )
+            await cloud_model_catalog_service.initialize()
+            ap.cloud_model_catalog_service = cloud_model_catalog_service
+
         ap.workspace_collaboration_service = workspace_collaboration_module.WorkspaceCollaborationService(
             ap,
             workspace_service_inst,

src/langbot/pkg/core/stages/load_config.py

@@ -41,6 +41,7 @@ _RUNTIME_POLICY_DEFAULTS = {
         }
     },
     'plugin': {
+        'connect_timeout_seconds': 180.0,
         'worker': {
             'max_cpus': 1.0,
             'max_memory_mb': 512,
@@ -56,7 +57,7 @@ _RUNTIME_POLICY_DEFAULTS = {
             'restart_failure_window_seconds': 30.0,
             'restart_circuit_open_seconds': 60.0,
             'require_hard_limits': False,
-        }
+        },
     },
     'mcp': {'stdio': {'enabled': True}},
     'monitoring': {

src/langbot/pkg/entity/dto/space_model.py

@@ -47,3 +47,10 @@ class SpaceModel(pydantic.BaseModel):
     status: str
     created_at: str | None = None
     updated_at: str | None = None
+
+
+class SpaceModelSelection(pydantic.BaseModel):
+    """Minimal model identity returned by the ranked selection endpoint."""
+
+    uuid: str
+    model_id: str

src/langbot/pkg/entity/persistence/model.py

@@ -48,6 +48,12 @@ class LLMModel(Base):
     provider_uuid = sqlalchemy.Column(sqlalchemy.String(255), nullable=False)
     abilities = sqlalchemy.Column(sqlalchemy.JSON, nullable=False, default=[])
     context_length = sqlalchemy.Column(sqlalchemy.Integer, nullable=True)
+    reasoning_config = sqlalchemy.Column(
+        sqlalchemy.JSON,
+        nullable=False,
+        default=lambda: {'level': 'provider_default'},
+        server_default=sqlalchemy.text('\'{"level":"provider_default"}\''),
+    )
     extra_args = sqlalchemy.Column(sqlalchemy.JSON, nullable=False, default={})
     prefered_ranking = sqlalchemy.Column(sqlalchemy.Integer, nullable=False, default=0)
     created_at = sqlalchemy.Column(sqlalchemy.DateTime, nullable=False, server_default=sqlalchemy.func.now())

src/langbot/pkg/entity/persistence/workspace.py

@@ -40,6 +40,11 @@ class MembershipStatus(enum.StrEnum):
     REMOVED = 'removed'
 
 
+class MembershipSource(enum.StrEnum):
+    LOCAL = 'local'
+    CLOUD_PROJECTION = 'cloud_projection'
+
+
 class InvitationStatus(enum.StrEnum):
     PENDING = 'pending'
     ACCEPTED = 'accepted'
@@ -151,6 +156,11 @@ class WorkspaceMembership(Base):
         nullable=True,
     )
     joined_at = sqlalchemy.Column(sqlalchemy.DateTime, nullable=True)
+    source = sqlalchemy.Column(
+        sqlalchemy.String(32),
+        nullable=False,
+        server_default=MembershipSource.LOCAL.value,
+    )
     projection_revision = sqlalchemy.Column(sqlalchemy.BigInteger, nullable=False, server_default='0')
     created_at = sqlalchemy.Column(sqlalchemy.DateTime, nullable=False, server_default=sqlalchemy.func.now())
     updated_at = sqlalchemy.Column(
@@ -163,6 +173,13 @@ class WorkspaceMembership(Base):
     __table_args__ = (
         sqlalchemy.UniqueConstraint('workspace_uuid', 'account_uuid', name='uq_workspace_membership_account'),
         sqlalchemy.Index('ix_workspace_memberships_account_status', 'account_uuid', 'status'),
+        sqlalchemy.Index(
+            'uq_workspace_memberships_one_active_owner',
+            'workspace_uuid',
+            unique=True,
+            sqlite_where=sqlalchemy.text("role = 'owner' AND status = 'active'"),
+            postgresql_where=sqlalchemy.text("role = 'owner' AND status = 'active'"),
+        ),
         sqlalchemy.CheckConstraint(
             "role IN ('owner', 'admin', 'developer', 'operator', 'viewer')",
             name='ck_workspace_memberships_role',
@@ -171,6 +188,10 @@ class WorkspaceMembership(Base):
             "status IN ('active', 'disabled', 'removed')",
             name='ck_workspace_memberships_status',
         ),
+        sqlalchemy.CheckConstraint(
+            "source IN ('local', 'cloud_projection')",
+            name='ck_workspace_memberships_source',
+        ),
     )
 
 

src/langbot/pkg/persistence/alembic/versions/0016_space_launch_replay.py

@@ -0,0 +1,57 @@
+"""add durable replay protection for signed Space launch assertions
+
+Revision ID: 0016_space_launch_replay
+Revises: 0015_cloud_core_collab
+Create Date: 2026-07-31
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = '0016_space_launch_replay'
+down_revision = '0015_cloud_core_collab'
+branch_labels = None
+depends_on = None
+
+_TABLE = 'space_launch_assertion_consumptions'
+_POLICY = 'langbot_directory_projection'
+_SETTING = "NULLIF(current_setting('langbot.directory_instance_uuid', true), '')"
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    if _TABLE not in set(sa.inspect(conn).get_table_names()):
+        op.create_table(
+            _TABLE,
+            sa.Column('instance_uuid', sa.String(255), nullable=False),
+            sa.Column('jti', sa.String(255), nullable=False),
+            sa.Column('expires_at', sa.DateTime(timezone=True), nullable=False),
+            sa.Column('consumed_at', sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False),
+            sa.PrimaryKeyConstraint('instance_uuid', 'jti'),
+        )
+        op.create_index(
+            'ix_space_launch_assertion_consumptions_expiry',
+            _TABLE,
+            ['instance_uuid', 'expires_at'],
+            unique=False,
+        )
+    if conn.dialect.name == 'postgresql':
+        table = conn.dialect.identifier_preparer.quote(_TABLE)
+        policy = conn.dialect.identifier_preparer.quote(_POLICY)
+        expression = f'instance_uuid::text = {_SETTING}'
+        op.execute(sa.text(f'ALTER TABLE {table} ENABLE ROW LEVEL SECURITY'))
+        op.execute(sa.text(f'ALTER TABLE {table} FORCE ROW LEVEL SECURITY'))
+        op.execute(sa.text(f'DROP POLICY IF EXISTS {policy} ON {table}'))
+        op.execute(
+            sa.text(
+                f'CREATE POLICY {policy} ON {table} AS PERMISSIVE FOR ALL TO PUBLIC '
+                f'USING ({expression}) WITH CHECK ({expression})'
+            )
+        )
+
+
+def downgrade() -> None:
+    if _TABLE in set(sa.inspect(op.get_bind()).get_table_names()):
+        op.drop_table(_TABLE)

src/langbot/pkg/persistence/alembic/versions/0018_add_llm_reasoning_config.py

@@ -0,0 +1,57 @@
+"""add llm reasoning config
+
+Revision ID: 0018_llm_reasoning_config
+Revises: 0017_oss_workspace_identity
+Create Date: 2026-07-27
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = '0018_llm_reasoning_config'
+down_revision = '0017_oss_workspace_identity'
+branch_labels = None
+depends_on = None
+
+
+_LLM_MODELS = sa.table(
+    'llm_models',
+    sa.column('reasoning_config', sa.JSON()),
+)
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'llm_models' not in inspector.get_table_names():
+        return
+
+    columns = {column['name'] for column in inspector.get_columns('llm_models')}
+    if 'reasoning_config' in columns:
+        return
+
+    op.add_column(
+        'llm_models',
+        sa.Column(
+            'reasoning_config',
+            sa.JSON(),
+            nullable=True,
+            server_default=sa.text('\'{"level":"provider_default"}\''),
+        ),
+    )
+    conn.execute(_LLM_MODELS.update().values(reasoning_config={'level': 'provider_default'}))
+    with op.batch_alter_table('llm_models') as batch_op:
+        batch_op.alter_column('reasoning_config', existing_type=sa.JSON(), nullable=False)
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'llm_models' not in inspector.get_table_names():
+        return
+    columns = {column['name'] for column in inspector.get_columns('llm_models')}
+    if 'reasoning_config' in columns:
+        with op.batch_alter_table('llm_models') as batch_op:
+            batch_op.drop_column('reasoning_config')

src/langbot/pkg/persistence/alembic/versions/0018_merge_launch_replay.py

@@ -0,0 +1,21 @@
+"""merge the published Space launch replay and main migration branches
+
+Revision ID: 0018_merge_launch_replay
+Revises: 0016_space_launch_replay, 0017_oss_workspace_identity
+Create Date: 2026-08-01
+"""
+
+from __future__ import annotations
+
+revision = '0018_merge_launch_replay'
+down_revision = ('0016_space_launch_replay', '0017_oss_workspace_identity')
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    pass
+
+
+def downgrade() -> None:
+    pass

src/langbot/pkg/persistence/alembic/versions/0019_single_workspace_owner.py

@@ -0,0 +1,80 @@
+"""enforce one active owner per Workspace
+
+Revision ID: 0019_single_workspace_owner
+Revises: 0018_merge_launch_replay
+Create Date: 2026-08-02
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = '0019_single_workspace_owner'
+down_revision = '0018_merge_launch_replay'
+branch_labels = None
+depends_on = None
+
+_INDEX_NAME = 'uq_workspace_memberships_one_active_owner'
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'workspace_memberships' not in inspector.get_table_names():
+        return
+
+    # Ownership transfer used to promote a second member without demoting the
+    # original owner. Preserve the Workspace creator where possible and demote
+    # every historical extra owner before installing the database invariant.
+    op.execute(
+        sa.text(
+            """
+            WITH ranked_owners AS (
+                SELECT membership.uuid,
+                       ROW_NUMBER() OVER (
+                           PARTITION BY membership.workspace_uuid
+                           ORDER BY
+                               CASE
+                                   WHEN membership.account_uuid = workspace.created_by_account_uuid THEN 0
+                                   ELSE 1
+                               END,
+                               COALESCE(membership.joined_at, membership.created_at),
+                               membership.uuid
+                       ) AS owner_rank
+                FROM workspace_memberships AS membership
+                JOIN workspaces AS workspace
+                  ON workspace.uuid = membership.workspace_uuid
+                WHERE membership.role = 'owner'
+                  AND membership.status = 'active'
+            )
+            UPDATE workspace_memberships
+               SET role = 'admin'
+             WHERE uuid IN (
+                 SELECT uuid
+                   FROM ranked_owners
+                  WHERE owner_rank > 1
+             )
+            """
+        )
+    )
+    # Fresh installations may already have this index because SQLAlchemy
+    # metadata is created before Alembic advances the revision marker.
+    op.execute(
+        sa.text(
+            'CREATE UNIQUE INDEX IF NOT EXISTS '
+            'uq_workspace_memberships_one_active_owner '
+            'ON workspace_memberships (workspace_uuid) '
+            "WHERE role = 'owner' AND status = 'active'"
+        )
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'workspace_memberships' not in inspector.get_table_names():
+        return
+    index_names = {index['name'] for index in inspector.get_indexes('workspace_memberships')}
+    if _INDEX_NAME in index_names:
+        op.drop_index(_INDEX_NAME, table_name='workspace_memberships')

src/langbot/pkg/persistence/alembic/versions/001a_pgvector_dimension_3072.py

@@ -0,0 +1,43 @@
+"""enable 3072-dimensional pgvector embeddings
+
+Revision ID: 001a_pgvector_dimension_3072
+Revises: 0019_single_workspace_owner
+Create Date: 2026-08-05
+"""
+
+from __future__ import annotations
+import sqlalchemy as sa
+from alembic import op
+
+revision = '001a_pgvector_dimension_3072'
+down_revision = '0019_single_workspace_owner'
+branch_labels = None
+depends_on = None
+_TABLE = 'langbot_vectors'
+_CHECK = 'ck_langbot_vectors_embedding_dimension_enabled'
+_INDEX = 'ix_langbot_vectors_hnsw_cosine_3072'
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name != 'postgresql' or _TABLE not in sa.inspect(conn).get_table_names():
+        return
+    op.drop_constraint(_CHECK, _TABLE, type_='check')
+    op.create_check_constraint(_CHECK, _TABLE, 'embedding_dimension IN (384, 512, 768, 1024, 1536, 3072)')
+    op.execute(
+        sa.text(
+            f'CREATE INDEX {_INDEX} ON {_TABLE} USING hnsw ((embedding::halfvec(3072)) halfvec_cosine_ops) WHERE embedding_dimension = 3072'
+        )
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name != 'postgresql' or _TABLE not in sa.inspect(conn).get_table_names():
+        return
+    count = conn.scalar(sa.text(f'SELECT COUNT(*) FROM {_TABLE} WHERE embedding_dimension = 3072'))
+    if count:
+        raise RuntimeError('Cannot disable 3072-dimensional pgvector while matching embeddings exist')
+    op.drop_index(_INDEX, table_name=_TABLE)
+    op.drop_constraint(_CHECK, _TABLE, type_='check')
+    op.create_check_constraint(_CHECK, _TABLE, 'embedding_dimension IN (384, 512, 768, 1024, 1536)')

src/langbot/pkg/persistence/alembic/versions/0020_workspace_membership_source.py

@@ -0,0 +1,49 @@
+"""add explicit Workspace membership source
+
+Revision ID: 0020_membership_source
+Revises: 001a_pgvector_dimension_3072
+Create Date: 2026-08-06
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = '0020_membership_source'
+down_revision = '001a_pgvector_dimension_3072'
+branch_labels = None
+depends_on = None
+
+_CONSTRAINT_NAME = 'ck_workspace_memberships_source'
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'workspace_memberships' not in inspector.get_table_names():
+        return
+    if 'source' in {column['name'] for column in inspector.get_columns('workspace_memberships')}:
+        return
+
+    # No durable historical field distinguishes Directory-created revision-zero
+    # rows from Core invitations. Protect every existing row; production can
+    # reclassify separately after UUIDs have been verified against Space.
+    with op.batch_alter_table('workspace_memberships') as batch_op:
+        batch_op.add_column(sa.Column('source', sa.String(length=32), nullable=False, server_default='local'))
+        batch_op.create_check_constraint(
+            _CONSTRAINT_NAME,
+            "source IN ('local', 'cloud_projection')",
+        )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    if 'workspace_memberships' not in inspector.get_table_names():
+        return
+    if 'source' not in {column['name'] for column in inspector.get_columns('workspace_memberships')}:
+        return
+    with op.batch_alter_table('workspace_memberships') as batch_op:
+        batch_op.drop_constraint(_CONSTRAINT_NAME, type_='check')
+        batch_op.drop_column('source')

src/langbot/pkg/persistence/alembic/versions/0021_merge_reasoning_config.py

@@ -0,0 +1,21 @@
+"""merge reasoning config with the main migration branch
+
+Revision ID: 0021_merge_reasoning_config
+Revises: 0020_membership_source, 0018_llm_reasoning_config
+Create Date: 2026-08-09
+"""
+
+from __future__ import annotations
+
+revision = '0021_merge_reasoning_config'
+down_revision = ('0020_membership_source', '0018_llm_reasoning_config')
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    pass
+
+
+def downgrade() -> None:
+    pass

src/langbot/pkg/persistence/mgr.py

@@ -98,7 +98,7 @@ _WORKSPACE_ALEMBIC_REVISION = '0009_workspace_tenancy'
 _RESOURCE_SCOPE_ALEMBIC_REVISION = '0010_scope_resources'
 _OSS_WORKSPACE_METADATA_KEY = 'oss_workspace_uuid'
 _RELEASE_MIGRATION_ADVISORY_LOCK_ID = 0x4C414E47424F5432
-_PGVECTOR_ALLOWED_DIMENSIONS = (384, 512, 768, 1024, 1536)
+_PGVECTOR_ALLOWED_DIMENSIONS = (384, 512, 768, 1024, 1536, 3072)
 _RUNTIME_SCHEMA = 'public'
 _ALEMBIC_RUNTIME_TABLE = 'alembic_version'
 _RUNTIME_TABLE_PRIVILEGES = frozenset({'SELECT', 'INSERT', 'UPDATE', 'DELETE'})
@@ -1356,14 +1356,16 @@ class PersistenceManager:
             index = by_index.get(index_name)
             index_definition = normalized(None if index is None else index['definition'])
             predicate = normalized(None if index is None else index['predicate'])
+            vector_type = 'halfvec' if dimension > 2000 else 'vector'
+            operator_class = f'{vector_type}_cosine_ops'
             if (
                 index is None
                 or index['access_method'] != 'hnsw'
                 or index['is_valid'] is not True
                 or index['is_ready'] is not True
-                or f'vector({dimension})' not in index_definition
-                or f'(embedding)::vector({dimension})' not in index_definition
-                or 'vector_cosine_ops' not in index_definition
+                or f'{vector_type}({dimension})' not in index_definition
+                or f'(embedding)::{vector_type}({dimension})' not in index_definition
+                or operator_class not in index_definition
                 or predicate.strip('() ') != f'embedding_dimension = {dimension}'
             ):
                 raise RuntimeError(f'PostgreSQL pgvector ANN index {index_name!r} is invalid')

src/langbot/pkg/persistence/tenant_uow.py

@@ -13,7 +13,7 @@ import typing
 import sqlalchemy
 import sqlalchemy.ext.asyncio as sqlalchemy_asyncio
 import sqlalchemy.orm as sqlalchemy_orm
-from pgvector.sqlalchemy import Vector
+from pgvector.sqlalchemy import HALFVEC, Vector
 from sqlalchemy.dialects.postgresql.dml import OnConflictDoNothing as PostgreSQLOnConflictDoNothing
 from sqlalchemy.dialects.postgresql.dml import OnConflictDoUpdate as PostgreSQLOnConflictDoUpdate
 from sqlalchemy.dialects.sqlite.dml import OnConflictDoNothing as SQLiteOnConflictDoNothing
@@ -209,7 +209,7 @@ _ALLOWED_SCOPED_BUILTIN_FUNCTION_TYPES = {
     'now': sqlalchemy.sql.functions.now,
     'sum': sqlalchemy.sql.functions.sum,
 }
-_ALLOWED_SCOPED_GENERIC_FUNCTIONS = frozenset({'length', 'nullif'})
+_ALLOWED_SCOPED_GENERIC_FUNCTIONS = frozenset({'date_trunc', 'length', 'nullif'})
 _ALLOWED_SCOPED_CUSTOM_OPERATORS = frozenset({'<=>'})
 _ALLOWED_SCOPED_STATEMENT_TYPES = (
     sqlalchemy.sql.dml.UpdateBase,
@@ -281,7 +281,7 @@ def _validate_scoped_sql_type(
         return
     seen.add(identity)
 
-    if type(sql_type) is Vector:
+    if type(sql_type) in {Vector, HALFVEC}:
         return
     if not type(sql_type).__module__.startswith('sqlalchemy.'):
         raise ScopedSessionTransactionError('TenantUnitOfWork does not allow custom SQL types in public statements')
@@ -462,7 +462,7 @@ def _validate_scoped_statement_call(args: tuple[typing.Any, ...], kwargs: dict[s
         if isinstance(element, sqlalchemy.sql.elements.BindParameter) and element.literal_execute:
             raise ScopedSessionTransactionError('TenantUnitOfWork does not allow literal-execute SQL parameters')
 
-        if isinstance(element, sqlalchemy.sql.elements.Cast) and type(element.type) is not Vector:
+        if isinstance(element, sqlalchemy.sql.elements.Cast) and type(element.type) not in {Vector, HALFVEC}:
             raise ScopedSessionTransactionError(
                 'TenantUnitOfWork only allows the trusted pgvector cast used by tenant vector search'
             )

src/langbot/pkg/pipeline/controller.py

@@ -132,9 +132,7 @@ class Controller:
 
                             break
 
-                    if selected_query:  # 找到了
-                        queries.remove(selected_query)
-                    else:  # 没找到 说明：没有请求 或者 所有query对应的session都已达到并发上限
+                    if not selected_query:  # No query is runnable under the current session limits.
                         await self.ap.query_pool.condition.wait()
                         continue
 

src/langbot/pkg/pipeline/preproc/preproc.py

@@ -41,6 +41,29 @@ class PreProcessor(stage.PipelineStage):
         selected_tool_names = {tool for tool in selected_tools if isinstance(tool, str)}
         return [tool for tool in tools if tool.name in selected_tool_names]
 
+    @staticmethod
+    def _append_to_system_prompt(
+        messages: list[provider_message.Message],
+        addition: str,
+    ) -> None:
+        """Append text to the first system message, creating one if none exists.
+
+        Handles both plain-string and content-element (list) message bodies.
+        """
+        if messages and messages[0].role == 'system':
+            head = messages[0]
+            if isinstance(head.content, str):
+                head.content = head.content + addition
+            elif isinstance(head.content, list):
+                for ce in head.content:
+                    if getattr(ce, 'type', None) == 'text':
+                        ce.text = (ce.text or '') + addition
+                        break
+                else:
+                    head.content.append(provider_message.ContentElement(type='text', text=addition))
+        else:
+            messages.insert(0, provider_message.Message(role='system', content=addition.strip()))
+
     async def process(
         self,
         query: pipeline_query.Query,
@@ -275,6 +298,23 @@ class PreProcessor(stage.PipelineStage):
         query.prompt.messages = event_ctx.event.default_prompt
         query.messages = event_ctx.event.prompt
 
+        # =========== Current date grounding for the local-agent runner ===========
+        # local-agent system prompts are static strings with no template-variable
+        # support, so without an explicit anchor the LLM resolves relative time
+        # references (e.g. "this quarter", "latest", "currently") against whichever
+        # period is best represented in its training data instead of the real date,
+        # and won't reliably know to double check time-sensitive facts with a tool.
+        if selected_runner == 'local-agent':
+            date_addition = (
+                f'\n\nCurrent date: {datetime.datetime.now().strftime("%Y-%m-%d (%A)")}. '
+                'Resolve relative time references (e.g. "today", "this quarter", "latest", '
+                '"currently") based on this date, not your training cutoff. For anything '
+                'time-sensitive that may have changed since training — stock prices, '
+                'financial results, news, current events, exchange rates, or similar — '
+                'verify with a search tool if one is available rather than answering from memory.'
+            )
+            self._append_to_system_prompt(query.prompt.messages, date_addition)
+
         # =========== Skill awareness for the local-agent runner ===========
         # The actual activation goes through the ``activate`` Tool Call so the
         # LLM doesn't see full SKILL.md instructions until it commits to a
@@ -310,27 +350,7 @@ class PreProcessor(stage.PipelineStage):
                 bound_skills=bound_skills,
             )
             if skill_addition:
-                # Append to the first system message; create one if the
-                # prompt has none. Handles both plain-string and
-                # content-element (list) message bodies.
-                if query.prompt.messages and query.prompt.messages[0].role == 'system':
-                    head = query.prompt.messages[0]
-                    if isinstance(head.content, str):
-                        head.content = head.content + skill_addition
-                    elif isinstance(head.content, list):
-                        appended = False
-                        for ce in head.content:
-                            if getattr(ce, 'type', None) == 'text':
-                                ce.text = (ce.text or '') + skill_addition
-                                appended = True
-                                break
-                        if not appended:
-                            head.content.append(provider_message.ContentElement(type='text', text=skill_addition))
-                else:
-                    query.prompt.messages.insert(
-                        0,
-                        provider_message.Message(role='system', content=skill_addition.strip()),
-                    )
+                self._append_to_system_prompt(query.prompt.messages, skill_addition)
                 self.ap.logger.debug(
                     f'Skill index injected into system prompt: '
                     f'pipeline={query.pipeline_uuid} '

src/langbot/pkg/platform/sources/websocket_adapter.py

@@ -5,6 +5,7 @@ import contextvars
 import logging
 import time
 import typing
+from dataclasses import dataclass
 from datetime import datetime
 
 import pydantic
@@ -25,6 +26,15 @@ _current_pipeline_uuid: contextvars.ContextVar[str | None] = contextvars.Context
 )
 
 
+@dataclass(frozen=True)
+class WebSocketReplyContext:
+    """Trusted routing context retained when the originating socket reconnects."""
+
+    scope: WebSocketScope
+    pipeline_uuid: str
+    session_id: str | None
+
+
 class WebSocketMessage(pydantic.BaseModel):
     """WebSocket消息格式"""
 
@@ -265,6 +275,11 @@ class WebSocketAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter)
         embed_target = self._parse_embed_target(sender_id)
         if embed_target is not None:
             return embed_target
+        reply_context = getattr(message_source, '_websocket_reply_context', None)
+        if isinstance(reply_context, WebSocketReplyContext):
+            if reply_context.scope != self._scope():
+                raise ValueError('WebSocket reply context does not match this adapter scope')
+            return reply_context.pipeline_uuid, reply_context.session_id
         raise ValueError('WebSocket reply target is not bound to this adapter scope')
 
     async def send_message(
@@ -685,6 +700,16 @@ class WebSocketAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter)
 
         # 异步触发事件处理
         # Use owner_bot's listeners if available, otherwise fall back to proxy bot
+        object.__setattr__(
+            event,
+            '_websocket_reply_context',
+            WebSocketReplyContext(
+                scope=connection.scope,
+                pipeline_uuid=pipeline_uuid,
+                session_id=connection.session_id,
+            ),
+        )
+
         listeners = (
             owner_bot.adapter.listeners
             if (owner_bot and hasattr(owner_bot.adapter, 'listeners') and owner_bot.adapter.listeners)
@@ -707,28 +732,37 @@ class WebSocketAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter)
             if len(listener_tasks) >= 100:
                 await self.logger.warning('WebSocket inbound listener capacity reached; dropping message')
                 return
-            token = _current_pipeline_uuid.set(pipeline_uuid)
-            try:
-                task_manager = getattr(self.ap, 'task_mgr', None)
-                if task_manager is None or not isinstance(getattr(task_manager, 'tasks', None), list):
-                    listener_task = asyncio.create_task(listeners[event.__class__](event, callback_adapter))
-                else:
-                    listener_task = task_manager.create_task(
-                        listeners[event.__class__](event, callback_adapter),
-                        kind='websocket-message',
-                        name=f'websocket-message-{connection.connection_id}',
-                        scopes=[
-                            core_entities.LifecycleControlScope.APPLICATION,
-                            core_entities.LifecycleControlScope.PLATFORM,
-                        ],
-                        instance_uuid=connection.instance_uuid,
-                        workspace_uuid=connection.workspace_uuid,
-                        placement_generation=connection.placement_generation,
-                    ).task
-                listener_tasks.add(listener_task)
-                listener_task.add_done_callback(self._listener_task_done)
-            finally:
-                _current_pipeline_uuid.reset(token)
+            listener = typing.cast(
+                typing.Callable[[typing.Any, typing.Any], typing.Awaitable[None]],
+                listeners[event.__class__],
+            )
+
+            async def run_listener():
+                token = _current_pipeline_uuid.set(pipeline_uuid)
+                try:
+                    await listener(event, callback_adapter)
+                finally:
+                    _current_pipeline_uuid.reset(token)
+
+            listener_coro = run_listener()
+            task_manager = getattr(self.ap, 'task_mgr', None)
+            if task_manager is None or not isinstance(getattr(task_manager, 'tasks', None), list):
+                listener_task = asyncio.create_task(listener_coro)
+            else:
+                listener_task = task_manager.create_task(
+                    listener_coro,
+                    kind='websocket-message',
+                    name=f'websocket-message-{connection.connection_id}',
+                    scopes=[
+                        core_entities.LifecycleControlScope.APPLICATION,
+                        core_entities.LifecycleControlScope.PLATFORM,
+                    ],
+                    instance_uuid=connection.instance_uuid,
+                    workspace_uuid=connection.workspace_uuid,
+                    placement_generation=connection.placement_generation,
+                ).task
+            listener_tasks.add(listener_task)
+            listener_task.add_done_callback(self._listener_task_done)
 
     def get_websocket_messages(
         self,

src/langbot/pkg/plugin/connector.py

@@ -6,6 +6,7 @@ import contextlib
 import contextvars
 import hashlib
 import json
+import math
 import time
 import uuid
 from typing import Any
@@ -76,7 +77,7 @@ _GITHUB_ASSET_HOSTS = frozenset(
     }
 )
 _HTTP_REDIRECT_STATUSES = frozenset({301, 302, 303, 307, 308})
-_CONNECT_TIMEOUT_SEC = 30.0
+_DEFAULT_CONNECT_TIMEOUT_SECONDS = 180.0
 _HEARTBEAT_INTERVAL_SEC = 20.0
 _HEARTBEAT_FAILURE_THRESHOLD = 3
 _RECONNECT_MAX_DELAY_SEC = 60.0
@@ -206,6 +207,17 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
 
         return f'{constants.instance_id}:plugin-runtime'
 
+    @staticmethod
+    def _runtime_connect_timeout(plugin_config: dict[str, Any]) -> float:
+        value = plugin_config.get('connect_timeout_seconds', _DEFAULT_CONNECT_TIMEOUT_SECONDS)
+        if isinstance(value, bool) or not isinstance(value, (int, float)) or not math.isfinite(value) or value <= 0:
+            raise ValueError('plugin.connect_timeout_seconds must be a positive number')
+        return float(value)
+
+    @staticmethod
+    def _runtime_connect_timeout_error(timeout_seconds: float) -> str:
+        return f'Plugin runtime did not become ready within {timeout_seconds:g} seconds'
+
     def _runtime_handler(self) -> handler.RuntimeConnectionHandler:
         runtime_handler = getattr(self, 'handler', None)
         if runtime_handler is None:
@@ -251,6 +263,8 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
     def _control_headers(self, *, allow_generate: bool) -> dict[str, str]:
         if not self._control_token and allow_generate:
             self._control_token = secrets.token_urlsafe(48)
+        if not self._control_token:
+            return {}
         try:
             self._control_token = validate_runtime_secret(
                 self._control_token,
@@ -699,10 +713,13 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
         """
 
         runtime_handler = self._runtime_handler()
+        started_at = time.monotonic()
         async with self._state_lock:
             all_states: dict[str, PluginInstallationDesiredState] = {}
             workspace_installations: dict[str, set[str]] = {}
+            workspace_count = 0
             for context in contexts:
+                workspace_count += 1
                 execution_context = await self._validate_execution_context(context)
                 states = await self._load_workspace_desired_states(execution_context)
                 installation_ids = {state.binding.installation_uuid for state in states}
@@ -722,6 +739,13 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
                     runtime_handler.unregister_installation_binding(previous.binding)
             self._known_desired_states = all_states
             self._workspace_installations = workspace_installations
+            self.ap.logger.info(
+                'Shared plugin runtime reconcile completed: workspaces=%d desired_installations=%d '
+                'elapsed_seconds=%.3f',
+                workspace_count,
+                len(all_states),
+                time.monotonic() - started_at,
+            )
             return result
 
     async def _validate_execution_context(self, context: TenantContext) -> ExecutionContext:
@@ -817,6 +841,8 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
             runtime_id=self._runtime_id,
         )
         self.worker_policy = self._load_worker_policy()
+        plugin_config = self.ap.instance_config.data.get('plugin', {})
+        connect_timeout_seconds = self._runtime_connect_timeout(plugin_config)
 
         async with self._lifecycle_lock:
             if self._closing:
@@ -958,10 +984,12 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
 
             self._transport_task = asyncio.create_task(task_coro)
             try:
-                await asyncio.wait_for(self._connected.wait(), timeout=_CONNECT_TIMEOUT_SEC)
+                await asyncio.wait_for(self._connected.wait(), timeout=connect_timeout_seconds)
             except asyncio.TimeoutError as exc:
                 await self._stop_transport()
-                raise PluginRuntimeNotConnectedError('Plugin runtime did not become ready within 30 seconds') from exc
+                raise PluginRuntimeNotConnectedError(
+                    self._runtime_connect_timeout_error(connect_timeout_seconds)
+                ) from exc
             if connect_errors:
                 await self._stop_transport()
                 raise PluginRuntimeNotConnectedError(f'Plugin runtime connection failed: {connect_errors[-1]}')
@@ -1968,11 +1996,11 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
         with runtime_handler.installation_scope(binding):
             return await runtime_handler.handle_page_api(plugin_author, plugin_name, page_id, endpoint, method, body)
 
-    async def get_debug_info(self) -> dict[str, Any]:
+    async def get_debug_info(self, execution_context: ExecutionContext) -> dict[str, Any]:
         """Get debug information including debug key and WS URL"""
         if not self.is_enable_plugin or not self._runtime_available():
             return {}
-        return await self._runtime_handler().get_debug_info()
+        return await self._runtime_handler().get_debug_info(execution_context)
 
     async def emit_event(
         self,

src/langbot/pkg/plugin/handler.py

@@ -1960,14 +1960,14 @@ class RuntimeConnectionHandler(handler.Handler):
         )
         return result
 
-    async def get_debug_info(self) -> dict[str, Any]:
+    async def get_debug_info(self, execution_context: ExecutionContext) -> dict[str, Any]:
         """Get debug information including debug key and WS URL"""
-        with self.installation_scope(None):
-            result = await self.call_action(
-                LangBotToRuntimeAction.GET_DEBUG_INFO,
-                {},
-                timeout=10,
-            )
+        result = await self.call_action(
+            LangBotToRuntimeAction.GET_DEBUG_INFO,
+            {},
+            timeout=10,
+            action_context=execution_context,
+        )
         return result
 
     # ================= RAG Capability Callers (LangBot -> Runtime) =================

src/langbot/pkg/provider/modelmgr/modelmgr.py

@@ -649,6 +649,7 @@ class ModelManager:
             provider_uuid=runtime_provider.provider_entity.uuid,
             abilities=model_info.get('abilities', []),
             context_length=model_info.get('context_length'),
+            reasoning_config=model_info.get('reasoning_config', {'level': 'provider_default'}),
             extra_args=model_info.get('extra_args', {}),
         )
         return self._build_llm_model(execution_context, model_entity, runtime_provider)
@@ -717,7 +718,10 @@ class ModelManager:
         provider_entity = self._coerce_provider(provider_info, context)
         requester_manifest = self.get_available_requester_manifest_by_name(provider_entity.requester)
         litellm_provider = self._get_litellm_provider_from_manifest(requester_manifest)
-        config = {'base_url': provider_entity.base_url}
+        config = {
+            'base_url': provider_entity.base_url,
+            'requester_name': provider_entity.requester,
+        }
 
         if litellm_provider:
             from .requesters import litellmchat

src/langbot/pkg/provider/modelmgr/reasoning.py

@@ -0,0 +1,125 @@
+from __future__ import annotations
+
+import typing
+
+
+ReasoningLevel = typing.Literal[
+    'provider_default',
+    'disabled',
+    'enabled',
+    'minimal',
+    'low',
+    'medium',
+    'high',
+    'xhigh',
+    'max',
+]
+
+REASONING_LEVELS: tuple[str, ...] = (
+    'provider_default',
+    'disabled',
+    'enabled',
+    'minimal',
+    'low',
+    'medium',
+    'high',
+    'xhigh',
+    'max',
+)
+DEFAULT_REASONING_CONFIG: dict[str, str] = {'level': 'provider_default'}
+
+_CONFLICTING_TOP_LEVEL_ARGS = {
+    'reasoning_effort',
+    'thinking',
+    'enable_thinking',
+    'thinking_budget',
+    'reasoning',
+}
+_CONFLICTING_EXTRA_BODY_ARGS = {
+    'reasoning_effort',
+    'thinking',
+    'enable_thinking',
+    'thinking_budget',
+    'reasoning',
+}
+
+
+def normalize_reasoning_config(value: typing.Any) -> dict[str, str]:
+    """Return the canonical model reasoning configuration."""
+    if value is None:
+        return dict(DEFAULT_REASONING_CONFIG)
+    if not isinstance(value, dict):
+        raise ValueError('reasoning_config must be an object')
+
+    unknown_fields = set(value) - {'level'}
+    if unknown_fields:
+        raise ValueError(f'Unsupported reasoning_config fields: {", ".join(sorted(unknown_fields))}')
+
+    level = value.get('level', 'provider_default')
+    if level not in REASONING_LEVELS:
+        raise ValueError(f'Unsupported reasoning level: {level}')
+    return {'level': typing.cast(str, level)}
+
+
+def validate_reasoning_config(
+    value: typing.Any,
+    abilities: typing.Iterable[str] | None,
+    extra_args: typing.Any,
+) -> dict[str, str]:
+    """Validate a model-facing reasoning config and conflicting raw arguments."""
+    config = normalize_reasoning_config(value)
+    if config['level'] == 'provider_default':
+        return config
+
+    if 'reasoning' not in set(abilities or []):
+        raise ValueError('The reasoning ability must be enabled before selecting a reasoning level')
+
+    conflicts = find_reasoning_arg_conflicts(extra_args)
+    if conflicts:
+        raise ValueError('reasoning_config conflicts with advanced parameters: ' + ', '.join(conflicts))
+    return config
+
+
+def find_reasoning_arg_conflicts(extra_args: typing.Any) -> list[str]:
+    if not isinstance(extra_args, dict):
+        return []
+
+    conflicts = [key for key in sorted(_CONFLICTING_TOP_LEVEL_ARGS) if key in extra_args]
+    extra_body = extra_args.get('extra_body')
+    if isinstance(extra_body, dict):
+        conflicts.extend(f'extra_body.{key}' for key in sorted(_CONFLICTING_EXTRA_BODY_ARGS) if key in extra_body)
+    return conflicts
+
+
+def validate_reasoning_capabilities(
+    config: typing.Any,
+    capabilities: typing.Mapping[str, typing.Any],
+    model_name: str,
+) -> None:
+    """Ensure an explicit reasoning level can be honored by the requester."""
+    level = normalize_reasoning_config(config)['level']
+    if level == 'provider_default':
+        return
+
+    available_levels = capabilities.get('levels')
+    if not isinstance(available_levels, list):
+        available_levels = []
+    legacy_levels = capabilities.get('legacy_levels')
+    if not isinstance(legacy_levels, list):
+        legacy_levels = []
+    if capabilities.get('supported') is not True or (level not in available_levels and level not in legacy_levels):
+        available_text = ', '.join(str(item) for item in available_levels) or 'provider_default'
+        raise ValueError(
+            f'Reasoning level "{level}" is not supported by model {model_name}. Available levels: {available_text}'
+        )
+
+
+def default_reasoning_capabilities(
+    supported: bool = False,
+    source: str = 'unknown',
+) -> dict[str, typing.Any]:
+    return {
+        'supported': supported,
+        'levels': ['provider_default'],
+        'source': source,
+    }

src/langbot/pkg/provider/modelmgr/requester.py

@@ -10,6 +10,7 @@ from ...entity.persistence import model as persistence_model
 from ...workspace.errors import WorkspaceInvariantError
 import langbot_plugin.api.entities.builtin.resource.tool as resource_tool
 from . import token
+from . import reasoning
 import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
 
@@ -377,11 +378,15 @@ class RuntimeLLMModel:
     provider: RuntimeProvider
     """提供商实例"""
 
+    reasoning_config_override: dict[str, str] | None
+    """Request-scoped reasoning policy supplied by the active pipeline."""
+
     def __init__(
         self,
         execution_context: ExecutionContext,
         model_entity: persistence_model.LLMModel,
         provider: RuntimeProvider,
+        reasoning_config_override: dict[str, str] | None = None,
     ):
         _ensure_same_execution_scope(provider.execution_context, execution_context, resource='LLM model')
         if model_entity.workspace_uuid != execution_context.workspace_uuid:
@@ -391,6 +396,7 @@ class RuntimeLLMModel:
         self.execution_context = execution_context
         self.model_entity = model_entity
         self.provider = provider
+        self.reasoning_config_override = reasoning_config_override
 
 
 class RuntimeEmbeddingModel:
@@ -482,6 +488,13 @@ class ProviderAPIRequester(metaclass=abc.ABCMeta):
         """
         raise NotImplementedError('This provider does not support model scanning')
 
+    def get_reasoning_capabilities(self, model: RuntimeLLMModel) -> dict[str, typing.Any]:
+        """Return normalized reasoning controls supported by a model."""
+        return reasoning.default_reasoning_capabilities(
+            supported='reasoning' in (model.model_entity.abilities or []),
+            source='manual' if 'reasoning' in (model.model_entity.abilities or []) else 'unknown',
+        )
+
     @abc.abstractmethod
     async def invoke_llm(
         self,

src/langbot/pkg/provider/modelmgr/requesters/litellmchat.py

@@ -7,7 +7,7 @@ import typing
 import litellm
 from litellm import acompletion, aembedding, arerank
 
-from .. import errors, requester
+from .. import errors, reasoning, requester
 from ....utils import httpclient
 import langbot_plugin.api.entities.builtin.resource.tool as resource_tool
 import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
@@ -164,6 +164,39 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
 
     _EMBEDDING_MODEL_HINTS = ('embedding', 'embed', 'bge-', 'e5-', 'm3e', 'gte-', 'text-embedding')
     _RERANK_MODEL_HINTS = ('rerank', 're-rank', 're_rank')
+    _QWEN_DEDICATED_THINKING_MODELS = frozenset(
+        {
+            'qwen3.7-max-preview',
+            'qwen3.7-max-2026-05-17',
+        }
+    )
+    _QWEN_REASONING_BUDGETS = {
+        'low': 1024,
+        'medium': 4096,
+        'high': 8192,
+    }
+    _INFERRED_EFFORT_PROVIDERS = frozenset(
+        {
+            'anthropic',
+            'gemini',
+            'groq',
+            'mistral',
+            'openai',
+            'openrouter',
+            'together_ai',
+            'xai',
+        }
+    )
+    _REQUESTER_REASONING_FAMILIES = {
+        'openai-chat-completions': 'openai',
+        'anthropic-messages': 'anthropic',
+        'deepseek-chat-completions': 'deepseek',
+        'moonshot-chat-completions': 'kimi',
+        'moonshot-cn-chat-completions': 'kimi',
+        'bailian-chat-completions': 'qwen',
+        'doubao-chat-completions': 'doubao',
+        'mimo-chat-completions': 'mimo',
+    }
 
     default_config: dict[str, typing.Any] = {
         'base_url': '',
@@ -172,6 +205,7 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
         'drop_params': False,
         'num_retries': 0,
         'api_version': '',
+        'requester_name': '',
     }
 
     async def initialize(self):
@@ -201,7 +235,10 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
             return False
 
         provider = self._get_custom_llm_provider()
-        candidates: list[tuple[str, str | None]] = [(model_name, provider)]
+        candidates: list[tuple[str, str | None]] = [
+            (candidate, None) for candidate in self._metadata_model_candidates(model_name)
+        ]
+        candidates.append((model_name, provider))
         litellm_model_name = self._build_litellm_model_name(model_name)
         if litellm_model_name != model_name:
             candidates.append((litellm_model_name, None))
@@ -268,6 +305,14 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
                 deduped_candidates.append(candidate)
         return deduped_candidates
 
+    @staticmethod
+    def _metadata_model_candidates(model_name: str) -> list[str]:
+        """Return known equivalent model IDs used only for LiteLLM metadata lookup."""
+        normalized_model_name = (model_name or '').lower()
+        if normalized_model_name.startswith('mimo-v2.5'):
+            return [f'openrouter/xiaomi/{normalized_model_name}']
+        return []
+
     def _known_context_length_fallback(self, model_name: str) -> int | None:
         normalized_model_name = (model_name or '').lower()
         if normalized_model_name.startswith('deepseek-v4-'):
@@ -287,7 +332,8 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
         if not callable(helper):
             return self._known_context_length_fallback(model_name)
 
-        candidates = [model_name]
+        candidates = self._metadata_model_candidates(model_name)
+        candidates.append(model_name)
         litellm_model_name = self._build_litellm_model_name(model_name)
         if litellm_model_name != model_name:
             candidates.append(litellm_model_name)
@@ -314,6 +360,297 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
     def _supports_vision(self, model_name: str) -> bool:
         return self._safe_litellm_bool_helper('supports_vision', model_name)
 
+    def _supports_reasoning(self, model_name: str) -> bool:
+        return self._safe_litellm_bool_helper('supports_reasoning', model_name)
+
+    def _requester_name(self, model: requester.RuntimeLLMModel | None = None) -> str:
+        if model is not None:
+            provider_entity = getattr(getattr(model, 'provider', None), 'provider_entity', None)
+            name = getattr(provider_entity, 'requester', None)
+            if isinstance(name, str) and name:
+                return name.lower()
+        return str(self.requester_cfg.get('requester_name') or '').lower()
+
+    @staticmethod
+    def _infer_reasoning_family_from_model_name(model_name: str) -> str:
+        normalized_name = (model_name or '').lower()
+        basename = normalized_name.rsplit('/', 1)[-1]
+        if basename.startswith(('gpt-', 'chatgpt-', 'o1', 'o3', 'o4')):
+            return 'openai'
+        if basename.startswith('claude-'):
+            return 'anthropic'
+        if basename.startswith('deepseek-'):
+            return 'deepseek'
+        if basename.startswith(('kimi-', 'moonshot-')):
+            return 'kimi'
+        if basename.startswith(('qwen-', 'qwen3', 'qwq')):
+            return 'qwen'
+        if basename.startswith(('doubao-', 'seed-')):
+            return 'doubao'
+        if basename.startswith('mimo-'):
+            return 'mimo'
+        return ''
+
+    def _reasoning_family(
+        self,
+        model_name: str,
+        model: requester.RuntimeLLMModel | None = None,
+    ) -> str:
+        requester_name = self._requester_name(model)
+        if requester_name in {'new-api-chat-completions', 'volcark-chat-completions'}:
+            inferred_family = self._infer_reasoning_family_from_model_name(model_name)
+            if inferred_family:
+                return inferred_family
+            return 'volcengine' if requester_name == 'volcark-chat-completions' else ''
+
+        # Bailian's compatible endpoint also hosts Kimi models. Keep those
+        # models on Kimi's ``thinking`` protocol instead of Qwen's
+        # ``enable_thinking`` protocol.
+        if requester_name == 'bailian-chat-completions':
+            inferred_family = self._infer_reasoning_family_from_model_name(model_name)
+            if inferred_family == 'kimi':
+                return inferred_family
+
+        requester_family = self._REQUESTER_REASONING_FAMILIES.get(requester_name)
+        if requester_family:
+            return requester_family
+
+        inferred_family = self._infer_reasoning_family_from_model_name(model_name)
+        provider = (self._get_custom_llm_provider() or '').lower()
+        if provider == 'openai':
+            return inferred_family or ('openai' if requester_name in {'', 'openai'} else '')
+        if provider:
+            return provider
+        return inferred_family
+
+    @staticmethod
+    def _is_anthropic_adaptive_model(model_name: str) -> bool:
+        basename = model_name.lower().rsplit('/', 1)[-1]
+        if 'mythos-preview' in basename:
+            return True
+
+        parts = basename.split('-')
+        if len(parts) < 3 or parts[0] != 'claude':
+            return False
+        model_families = {'opus', 'sonnet', 'fable', 'mythos'}
+        if parts[1] in model_families:
+            if parts[2] == '5':
+                return True
+            return len(parts) >= 4 and parts[2] == '4' and parts[3] in {'6', '7', '8'}
+        return parts[1] == '5' and parts[2] in model_families
+
+    @staticmethod
+    def _is_anthropic_always_thinking_model(model_name: str) -> bool:
+        normalized_name = model_name.lower()
+        return any(marker in normalized_name for marker in ('fable-5', 'mythos-5', 'mythos-preview'))
+
+    @staticmethod
+    def _is_dedicated_qwen_thinking_model(model_name: str) -> bool:
+        normalized_name = model_name.lower().rsplit('/', 1)[-1]
+        return (
+            normalized_name in LiteLLMRequester._QWEN_DEDICATED_THINKING_MODELS
+            or normalized_name.startswith('qwq')
+            or '-thinking' in normalized_name
+        )
+
+    @staticmethod
+    def _supports_qwen_thinking_budget(model_name: str) -> bool:
+        """Return whether the documented Qwen3 family supports thinking_budget."""
+        normalized_name = model_name.lower().rsplit('/', 1)[-1]
+        return normalized_name.startswith('qwen3')
+
+    def _known_reasoning_levels(self, model_name: str, family: str) -> list[str] | None:
+        normalized_name = model_name.lower().rsplit('/', 1)[-1]
+
+        if family == 'deepseek' and normalized_name.startswith('deepseek-'):
+            if normalized_name.startswith('deepseek-v4-'):
+                return ['provider_default', 'disabled', 'low', 'high', 'xhigh', 'max']
+            if 'reasoner' in normalized_name or '-r1' in normalized_name:
+                return ['provider_default']
+            return ['provider_default', 'disabled', 'enabled']
+
+        if family == 'kimi':
+            if normalized_name.startswith('kimi-k3'):
+                return ['provider_default', 'low', 'high', 'max']
+            if normalized_name.startswith('kimi-k2.7-code'):
+                return ['provider_default']
+            if normalized_name.startswith(('kimi-k2.5', 'kimi-k2.6')):
+                return ['provider_default', 'disabled', 'enabled']
+            if 'thinking' in normalized_name:
+                return ['provider_default']
+
+        if family == 'qwen' and normalized_name.startswith(('qwen-', 'qwen3', 'qwq')):
+            if self._is_dedicated_qwen_thinking_model(normalized_name):
+                if self._supports_qwen_thinking_budget(normalized_name):
+                    return ['provider_default', 'low', 'medium', 'high']
+                return ['provider_default']
+            if self._supports_qwen_thinking_budget(normalized_name):
+                return ['provider_default', 'disabled', 'low', 'medium', 'high']
+            return ['provider_default', 'disabled', 'enabled']
+
+        if family == 'doubao' and normalized_name.startswith(('doubao-', 'seed-')):
+            return ['provider_default', 'disabled', 'low', 'medium', 'high']
+
+        if family == 'mimo' and normalized_name.startswith(('mimo-v2.5',)):
+            return ['provider_default', 'disabled', 'enabled']
+
+        if family == 'anthropic' and normalized_name.startswith('claude-'):
+            levels = ['provider_default']
+            adaptive = self._is_anthropic_adaptive_model(normalized_name)
+            if adaptive and not self._is_anthropic_always_thinking_model(normalized_name):
+                levels.append('disabled')
+            levels.extend(['low', 'medium', 'high'])
+            if adaptive:
+                levels.extend(['xhigh', 'max'])
+            return levels
+
+        if family == 'openai' and normalized_name.startswith(('gpt-5', 'o1', 'o3', 'o4')):
+            return ['provider_default', 'low', 'medium', 'high']
+
+        return None
+
+    def _openai_reasoning_levels(self, model_name: str) -> list[str]:
+        model_info = self._safe_model_info(model_name)
+        levels = ['provider_default']
+        if model_info.get('supports_none_reasoning_effort') is True:
+            levels.append('disabled')
+        if model_info.get('supports_minimal_reasoning_effort') is True:
+            levels.append('minimal')
+        for level in ('low', 'medium', 'high'):
+            if model_info.get(f'supports_{level}_reasoning_effort') is not False:
+                levels.append(level)
+        for level in ('xhigh', 'max'):
+            if model_info.get(f'supports_{level}_reasoning_effort') is True:
+                levels.append(level)
+        return levels
+
+    def _safe_model_info(self, model_name: str) -> dict[str, typing.Any]:
+        helper = getattr(litellm, 'get_model_info', None)
+        if not callable(helper):
+            return {}
+
+        candidates = [
+            *self._metadata_model_candidates(model_name),
+            model_name,
+            self._build_litellm_model_name(model_name),
+        ]
+        for candidate in candidates:
+            try:
+                info = helper(candidate)
+            except Exception:
+                continue
+            if isinstance(info, dict):
+                return info
+            model_dump = getattr(info, 'model_dump', None)
+            if callable(model_dump):
+                try:
+                    dumped = model_dump()
+                    if isinstance(dumped, dict):
+                        return dumped
+                except Exception:
+                    continue
+        return {}
+
+    def get_reasoning_capabilities(self, model: requester.RuntimeLLMModel) -> dict[str, typing.Any]:
+        model_name = model.model_entity.name
+        abilities = model.model_entity.abilities or []
+        detected = self._supports_reasoning(model_name)
+        declared = 'reasoning' in abilities
+        family = self._reasoning_family(model_name, model)
+        known_levels = self._known_reasoning_levels(model_name, family)
+        supported = detected or declared or known_levels is not None
+        if not supported:
+            return reasoning.default_reasoning_capabilities()
+
+        normalized_name = model_name.lower()
+        if family == 'openai':
+            levels = self._openai_reasoning_levels(model_name)
+        elif known_levels is not None:
+            levels = known_levels
+        elif family == 'anthropic':
+            levels = ['provider_default', 'low', 'medium', 'high']
+        elif family in {'deepseek', 'qwen', 'mimo', 'volcengine'}:
+            levels = ['provider_default', 'disabled', 'enabled']
+        elif family == 'doubao':
+            levels = ['provider_default', 'disabled', 'low', 'medium', 'high']
+        elif family == 'ollama':
+            levels = ['provider_default']
+            levels.append('disabled')
+            if normalized_name.startswith('gpt-oss') or '/gpt-oss' in normalized_name:
+                levels.extend(['low', 'medium', 'high'])
+            else:
+                levels.append('enabled')
+        elif family in self._INFERRED_EFFORT_PROVIDERS:
+            levels = ['provider_default', 'low', 'medium', 'high']
+        else:
+            levels = ['provider_default']
+
+        capabilities = {
+            'supported': True,
+            'levels': list(dict.fromkeys(levels)),
+            'source': 'litellm' if detected else ('provider' if known_levels is not None else 'manual'),
+        }
+        if family == 'qwen' and 'disabled' in capabilities['levels'] and 'enabled' not in capabilities['levels']:
+            capabilities['legacy_levels'] = ['enabled']
+        return capabilities
+
+    def _build_reasoning_args(self, model: requester.RuntimeLLMModel) -> dict[str, typing.Any]:
+        level = self._reasoning_level(model)
+        if level == 'provider_default':
+            return {}
+
+        config = {'level': level}
+        capabilities = self.get_reasoning_capabilities(model)
+        try:
+            reasoning.validate_reasoning_capabilities(config, capabilities, model.model_entity.name)
+        except ValueError as exc:
+            raise errors.RequesterError(str(exc)) from exc
+
+        family = self._reasoning_family(model.model_entity.name, model)
+        if level == 'disabled':
+            if family in {'deepseek', 'kimi', 'mimo', 'doubao'}:
+                return {'extra_body': {'thinking': {'type': 'disabled'}}}
+            if family == 'qwen':
+                return {'extra_body': {'enable_thinking': False}}
+            if family == 'volcengine':
+                return {'extra_body': {'thinking': {'type': 'disabled'}}}
+            if family == 'anthropic':
+                return {'thinking': {'type': 'disabled'}}
+            return {'reasoning_effort': 'none'}
+        if level == 'enabled':
+            if family in {'deepseek', 'kimi', 'mimo', 'volcengine'}:
+                return {'extra_body': {'thinking': {'type': 'enabled'}}}
+            if family == 'qwen':
+                return {'extra_body': {'enable_thinking': True}}
+            return {'reasoning_effort': 'low'}
+        if family == 'qwen' and level in self._QWEN_REASONING_BUDGETS:
+            return {
+                'extra_body': {
+                    'enable_thinking': True,
+                    'thinking_budget': self._QWEN_REASONING_BUDGETS[level],
+                }
+            }
+        if family == 'deepseek':
+            return {
+                'extra_body': {
+                    'thinking': {'type': 'enabled'},
+                    'reasoning_effort': level,
+                }
+            }
+        return {'reasoning_effort': level}
+
+    @staticmethod
+    def _reasoning_config_value(model: requester.RuntimeLLMModel) -> typing.Any:
+        raw_config = getattr(model, 'reasoning_config_override', None)
+        if raw_config is None:
+            raw_config = getattr(model.model_entity, 'reasoning_config', None)
+        if not isinstance(raw_config, dict):
+            return None
+        return raw_config
+
+    def _reasoning_level(self, model: requester.RuntimeLLMModel) -> str:
+        return reasoning.normalize_reasoning_config(self._reasoning_config_value(model))['level']
+
     def _infer_model_type(self, model_id: str) -> str:
         normalized_id = (model_id or '').lower()
         if any(kw in normalized_id for kw in self._RERANK_MODEL_HINTS):
@@ -344,6 +681,13 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
             )
             if supports_provider_reported_vision or self._supports_vision(model_id):
                 abilities.append('vision')
+            supports_provider_reported_reasoning = bool(
+                model_payload and model_payload.get('supports_reasoning') is True
+            )
+            family = self._reasoning_family(model_id)
+            supports_known_reasoning = self._known_reasoning_levels(model_id, family) is not None
+            if supports_provider_reported_reasoning or supports_known_reasoning or self._supports_reasoning(model_id):
+                abilities.append('reasoning')
             scanned_model['abilities'] = abilities
 
             context_length = self._context_length_from_scan_payload(model_payload)
@@ -354,13 +698,51 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
 
         return scanned_model
 
-    def _convert_messages(self, messages: typing.List[provider_message.Message]) -> list[dict]:
+    def _convert_messages(
+        self,
+        messages: typing.List[provider_message.Message],
+        reasoning_family: str = '',
+        include_reasoning_context: bool = True,
+    ) -> list[dict]:
         """Convert LangBot messages to LiteLLM/OpenAI format."""
         req_messages = []
         for m in messages:
             msg_dict = m.dict(exclude_none=True)
             content = msg_dict.get('content')
 
+            if msg_dict.get('role') == 'assistant' and reasoning_family:
+                provider_fields = msg_dict.get('provider_specific_fields')
+                if isinstance(provider_fields, dict):
+                    cleaned_provider_fields = dict(provider_fields)
+                    reasoning_content = cleaned_provider_fields.pop('reasoning_content', None)
+                    thinking_blocks = cleaned_provider_fields.pop('thinking_blocks', None)
+
+                    # ``content`` is also used for the user-facing rendering.
+                    # Do not replay that rendered <think> wrapper alongside the
+                    # structured provider reasoning on the next request.
+                    if reasoning_content or thinking_blocks:
+                        content = msg_dict.get('content')
+                        if isinstance(content, str):
+                            msg_dict['content'] = self._strip_think(content)
+
+                    if include_reasoning_context:
+                        if reasoning_family == 'anthropic' and thinking_blocks:
+                            msg_dict['thinking_blocks'] = thinking_blocks
+                        elif reasoning_family in {
+                            'deepseek',
+                            'kimi',
+                            'qwen',
+                            'doubao',
+                            'mimo',
+                            'volcengine',
+                        } and isinstance(reasoning_content, str):
+                            msg_dict['reasoning_content'] = reasoning_content
+
+                    if cleaned_provider_fields:
+                        msg_dict['provider_specific_fields'] = cleaned_provider_fields
+                    else:
+                        msg_dict.pop('provider_specific_fields', None)
+
             if isinstance(content, list):
                 converted_parts = []
                 for part in content:
@@ -421,6 +803,52 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
 
         return content or ''
 
+    @staticmethod
+    def _thinking_blocks_text(thinking_blocks: typing.Any) -> str:
+        if not isinstance(thinking_blocks, list):
+            return ''
+        parts = []
+        for block in thinking_blocks:
+            if isinstance(block, dict):
+                text = block.get('thinking')
+            else:
+                text = getattr(block, 'thinking', None)
+            if isinstance(text, str) and text:
+                parts.append(text)
+        return ''.join(parts)
+
+    @classmethod
+    def _merge_thinking_blocks(
+        cls,
+        current: list[dict[str, typing.Any]],
+        incoming: typing.Any,
+    ) -> list[dict[str, typing.Any]]:
+        """Merge Anthropic thinking block fragments emitted by a stream."""
+        if not isinstance(incoming, list):
+            return current
+        merged = [dict(block) for block in current]
+        for raw_block in incoming:
+            block = cls._as_dict(raw_block)
+            if not block:
+                continue
+            block_type = block.get('type')
+            if block_type == 'redacted_thinking':
+                merged.append(block)
+                continue
+
+            text = block.get('thinking') if isinstance(block.get('thinking'), str) else ''
+            signature = block.get('signature')
+            if merged and merged[-1].get('type') == 'thinking' and not merged[-1].get('signature'):
+                merged[-1]['thinking'] = f'{merged[-1].get("thinking", "")}{text}'
+                if signature:
+                    merged[-1]['signature'] = signature
+            elif merged and signature and merged[-1].get('signature') == signature:
+                if text and text != merged[-1].get('thinking', ''):
+                    merged[-1]['thinking'] = f'{merged[-1].get("thinking", "")}{text}'
+            else:
+                merged.append(block)
+        return merged
+
     @staticmethod
     def _normalize_usage(usage: typing.Any) -> dict:
         """Normalize a LiteLLM/OpenAI usage object into a plain token dict.
@@ -651,7 +1079,13 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
         stream: bool = False,
     ) -> dict:
         """Build common completion arguments for invoke_llm and invoke_llm_stream."""
-        req_messages = self._convert_messages(messages)
+        reasoning_family = self._reasoning_family(model.model_entity.name, model)
+        reasoning_level = self._reasoning_level(model)
+        req_messages = self._convert_messages(
+            messages,
+            reasoning_family=reasoning_family,
+            include_reasoning_context=reasoning_level != 'disabled',
+        )
         model_name = self._build_litellm_model_name(model.model_entity.name)
         api_key = model.provider.token_mgr.get_token()
 
@@ -670,6 +1104,29 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
             args.update(model.model_entity.extra_args)
         args.update(extra_args)
 
+        reasoning_args = self._build_reasoning_args(model)
+        if reasoning_args:
+            conflicts = reasoning.find_reasoning_arg_conflicts(model.model_entity.extra_args)
+            conflicts.extend(reasoning.find_reasoning_arg_conflicts(extra_args))
+            if conflicts:
+                raise errors.RequesterError(
+                    'reasoning_config conflicts with advanced parameters: ' + ', '.join(dict.fromkeys(conflicts))
+                )
+            reasoning_extra_body = reasoning_args.get('extra_body')
+            if isinstance(reasoning_extra_body, dict):
+                existing_extra_body = args.get('extra_body') or {}
+                if not isinstance(existing_extra_body, dict):
+                    raise errors.RequesterError('extra_body must be an object')
+                args.update({key: value for key, value in reasoning_args.items() if key != 'extra_body'})
+                args['extra_body'] = {**existing_extra_body, **reasoning_extra_body}
+            else:
+                args.update(reasoning_args)
+            if 'reasoning_effort' in reasoning_args and self._get_custom_llm_provider() == 'openai':
+                allowed_openai_params = args.get('allowed_openai_params') or []
+                if not isinstance(allowed_openai_params, (list, tuple, set)):
+                    raise errors.RequesterError('allowed_openai_params must be an array')
+                args['allowed_openai_params'] = list(dict.fromkeys([*allowed_openai_params, 'reasoning_effort']))
+
         if funcs:
             tools = await self.ap.tool_mgr.generate_tools_for_openai(funcs)
             if tools:
@@ -699,10 +1156,21 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
 
             content = message_data.get('content', '')
             reasoning_content = message_data.get('reasoning_content', None)
-            message_data['content'] = self._process_thinking_content(content, reasoning_content, remove_think)
+            thinking_blocks = message_data.get('thinking_blocks')
+            if reasoning_content or thinking_blocks:
+                provider_fields = dict(message_data.get('provider_specific_fields') or {})
+                if reasoning_content:
+                    provider_fields['reasoning_content'] = reasoning_content
+                if thinking_blocks:
+                    provider_fields['thinking_blocks'] = thinking_blocks
+                message_data['provider_specific_fields'] = provider_fields
+            display_reasoning = reasoning_content or self._thinking_blocks_text(thinking_blocks) or None
+            message_data['content'] = self._process_thinking_content(content, display_reasoning, remove_think)
 
             if 'reasoning_content' in message_data:
                 del message_data['reasoning_content']
+            if 'thinking_blocks' in message_data:
+                del message_data['thinking_blocks']
 
             message = provider_message.Message(**message_data)
             usage_info = self._extract_usage(response)
@@ -728,6 +1196,9 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
         role = 'assistant'
         tool_call_state: dict[int, dict[str, typing.Any]] = {}
         think_state = _ThinkStripState() if remove_think else None
+        reasoning_started = False
+        reasoning_closed = False
+        thinking_blocks_state: list[dict[str, typing.Any]] = []
 
         try:
             response = await acompletion(**args)
@@ -758,28 +1229,63 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
                 if 'role' in delta and delta['role']:
                     role = delta['role']
 
-                delta_content = delta.get('content', '')
-                reasoning_content = delta.get('reasoning_content', '')
+                delta_content = delta.get('content') or ''
+                reasoning_content = delta.get('reasoning_content') or ''
+                provider_fields = dict(delta.get('provider_specific_fields') or {})
+                raw_thinking_blocks = delta.get('thinking_blocks')
+                if raw_thinking_blocks:
+                    thinking_blocks_state = self._merge_thinking_blocks(thinking_blocks_state, raw_thinking_blocks)
+                    provider_fields['thinking_blocks'] = thinking_blocks_state
+                thinking_blocks_text = self._thinking_blocks_text(raw_thinking_blocks)
+                display_reasoning_content = reasoning_content or thinking_blocks_text
 
                 # Handle reasoning_content based on remove_think flag
                 if reasoning_content:
+                    provider_fields['reasoning_content'] = reasoning_content
                     if remove_think:
-                        # Skip reasoning content when remove_think is True
-                        chunk_idx += 1
-                        continue
+                        delta_content = delta_content or None
                     else:
-                        # Use reasoning_content as the displayed content
-                        delta_content = reasoning_content
+                        # Stream explicit markers so downstream adapters and
+                        # the debug page see the same format as non-streaming
+                        # responses.
+                        if not reasoning_started:
+                            delta_content = '<think>\n'
+                            reasoning_started = True
+                        else:
+                            delta_content = ''
+                        delta_content += display_reasoning_content
+                        if delta.get('content'):
+                            delta_content += f'\n</think>\n{delta.get("content")}'
+                            reasoning_closed = True
+
+                elif display_reasoning_content:
+                    if remove_think:
+                        delta_content = delta_content or None
+                    else:
+                        if not reasoning_started:
+                            delta_content = '<think>\n'
+                            reasoning_started = True
+                        else:
+                            delta_content = ''
+                        delta_content += display_reasoning_content
+                        if delta.get('content'):
+                            delta_content += f'\n</think>\n{delta.get("content")}'
+                            reasoning_closed = True
+
+                elif delta_content and not remove_think and reasoning_started and not reasoning_closed:
+                    delta_content = f'\n</think>\n{delta_content}'
+                    reasoning_closed = True
+
+                if finish_reason and not remove_think and reasoning_started and not reasoning_closed:
+                    delta_content = f'{delta_content}\n</think>\n'
+                    reasoning_closed = True
 
                 if think_state is not None and delta_content:
                     delta_content = think_state.feed(delta_content)
-                    if not delta_content:
-                        chunk_idx += 1
-                        continue
 
                 tool_calls = self._normalize_stream_tool_calls(delta.get('tool_calls'), tool_call_state)
 
-                if chunk_idx == 0 and not delta_content and not tool_calls:
+                if not delta_content and not tool_calls and not provider_fields and not finish_reason:
                     chunk_idx += 1
                     continue
 
@@ -791,13 +1297,20 @@ class LiteLLMRequester(requester.ProviderAPIRequester):
                 }
 
                 # Preserve provider_specific_fields from delta (e.g., Gemini thought_signatures)
-                if delta.get('provider_specific_fields'):
-                    chunk_data['provider_specific_fields'] = delta['provider_specific_fields']
+                if provider_fields:
+                    chunk_data['provider_specific_fields'] = provider_fields
 
                 chunk_data = {k: v for k, v in chunk_data.items() if v is not None}
                 yield provider_message.MessageChunk(**chunk_data)
                 chunk_idx += 1
 
+            if reasoning_started and not reasoning_closed:
+                yield provider_message.MessageChunk(
+                    role=role,
+                    content='\n</think>\n',
+                    is_final=True,
+                )
+
             if think_state is not None:
                 pending_content = think_state.flush()
                 if pending_content:

src/langbot/pkg/provider/runners/localagent.py

@@ -6,6 +6,7 @@ import typing
 from .. import runner
 from ...telemetry import features as telemetry_features
 from ..modelmgr import requester as modelmgr_requester
+from ..modelmgr import reasoning as modelmgr_reasoning
 from ..tools.loaders.native import EXEC_TOOL_NAME
 import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
@@ -60,6 +61,7 @@ class _StreamAccumulator:
         self.msg_idx = 0
         self.accumulated_content = initial_content or ''
         self.last_role = 'assistant'
+        self.provider_specific_fields: dict[str, typing.Any] = {}
         self.msg_sequence = msg_sequence
         self.remove_think = remove_think
         self._think_state = None
@@ -90,10 +92,27 @@ class _StreamAccumulator:
                             name=tool_call.function.name if tool_call.function else '',
                             arguments='',
                         ),
+                        provider_specific_fields=(
+                            dict(tool_call.provider_specific_fields) if tool_call.provider_specific_fields else None
+                        ),
                     )
+                elif tool_call.provider_specific_fields:
+                    existing_fields = self.tool_calls_map[tool_call.id].provider_specific_fields or {}
+                    self.tool_calls_map[tool_call.id].provider_specific_fields = {
+                        **existing_fields,
+                        **tool_call.provider_specific_fields,
+                    }
                 if tool_call.function and tool_call.function.arguments:
                     self.tool_calls_map[tool_call.id].function.arguments += tool_call.function.arguments
 
+        if msg.provider_specific_fields:
+            for key, value in msg.provider_specific_fields.items():
+                if key == 'reasoning_content' and isinstance(value, str):
+                    previous = self.provider_specific_fields.get(key, '')
+                    self.provider_specific_fields[key] = f'{previous}{value}'
+                else:
+                    self.provider_specific_fields[key] = value
+
         if msg.is_final:
             self._flush_think_state()
 
@@ -103,6 +122,7 @@ class _StreamAccumulator:
                 role=self.last_role,
                 content=self._maybe_strip_think(self.accumulated_content),
                 tool_calls=list(self.tool_calls_map.values()) if (self.tool_calls_map and msg.is_final) else None,
+                provider_specific_fields=(self.provider_specific_fields or None) if msg.is_final else None,
                 is_final=msg.is_final,
                 msg_sequence=self.msg_sequence,
             )
@@ -115,6 +135,7 @@ class _StreamAccumulator:
             role=self.last_role,
             content=self._maybe_strip_think(self.accumulated_content),
             tool_calls=list(self.tool_calls_map.values()) if self.tool_calls_map else None,
+            provider_specific_fields=self.provider_specific_fields or None,
             msg_sequence=self.msg_sequence,
         )
 
@@ -233,9 +254,10 @@ class LocalAgentRunner(runner.RequestRunner):
                     execution_context,
                     query.use_llm_model_uuid,
                 )
-                candidates.append(primary)
             except ValueError:
                 self.ap.logger.warning(f'Primary model {query.use_llm_model_uuid} not found')
+            else:
+                candidates.append(LocalAgentRunner._apply_pipeline_reasoning_config(query, primary))
 
         # Fallback models
         fallback_uuids = (query.variables or {}).get('_fallback_model_uuids', [])
@@ -245,12 +267,31 @@ class LocalAgentRunner(runner.RequestRunner):
                     execution_context,
                     fb_uuid,
                 )
-                candidates.append(fb_model)
             except ValueError:
                 self.ap.logger.warning(f'Fallback model {fb_uuid} not found, skipping')
+            else:
+                candidates.append(LocalAgentRunner._apply_pipeline_reasoning_config(query, fb_model))
 
         return candidates
 
+    @staticmethod
+    def _apply_pipeline_reasoning_config(
+        query: pipeline_query.Query,
+        model: modelmgr_requester.RuntimeLLMModel,
+    ) -> modelmgr_requester.RuntimeLLMModel:
+        local_agent_config = query.pipeline_config.get('ai', {}).get('local-agent', {})
+        model_config = local_agent_config.get('model', {})
+        reasoning_by_model = model_config.get('reasoning', {}) if isinstance(model_config, dict) else {}
+        level = (
+            reasoning_by_model.get(model.model_entity.uuid, 'provider_default')
+            if isinstance(reasoning_by_model, dict)
+            else 'provider_default'
+        )
+        reasoning_config = modelmgr_reasoning.normalize_reasoning_config({'level': level})
+        configured_model = copy.copy(model)
+        configured_model.reasoning_config_override = reasoning_config
+        return configured_model
+
     async def _invoke_with_fallback(
         self,
         query: pipeline_query.Query,

src/langbot/pkg/telemetry/heartbeat.py

@@ -37,6 +37,7 @@ class WorkspaceResourceSnapshot(typing.TypedDict):
     extension_count: int
     skill_count: int
     adapters: list[str]
+    execution_generation: int
 
 
 async def _count(
@@ -81,6 +82,7 @@ async def _cloud_workspace_resource_counts(ap: core_app.Application, bindings) -
             'extension_count': 0,
             'skill_count': 0,
             'adapters': [],
+            'execution_generation': binding.placement_generation,
         }
         for binding in bindings
     }
@@ -118,6 +120,7 @@ async def build_heartbeat_payload(
     ap: core_app.Application,
     *,
     workspace_uuid: str,
+    workspace_create_ts: int = 0,
     workspace_resource: WorkspaceResourceSnapshot | None = None,
 ) -> dict:
     """Collect one anonymous Workspace profile snapshot."""
@@ -210,7 +213,9 @@ async def build_heartbeat_payload(
         'event_type': 'instance_heartbeat',
         'query_id': '',
         'version': constants.semantic_version,
+        'instance_id': constants.instance_id,
         'workspace_uuid': workspace_uuid,
+        'workspace_create_ts': workspace_create_ts,
         'instance_create_ts': constants.instance_create_ts,
         'edition': constants.edition,
         'features': features,
@@ -218,10 +223,24 @@ async def build_heartbeat_payload(
     }
 
 
+def _workspace_created_timestamp(created_at: datetime | None) -> int:
+    if created_at is None:
+        return 0
+    if created_at.tzinfo is None:
+        # SQLAlchemy may return persisted UTC values without tzinfo. Never
+        # reinterpret them in the host's local timezone.
+        created_at = created_at.replace(tzinfo=timezone.utc)
+    return int(created_at.timestamp())
+
+
 async def build_heartbeat_payloads(ap: core_app.Application) -> list[dict]:
     """Build one heartbeat per active Workspace."""
     bindings = await ap.workspace_service.list_active_execution_bindings()
     workspace_uuids = sorted({binding.workspace_uuid for binding in bindings})
+    workspace_create_ts = {
+        binding.workspace_uuid: _workspace_created_timestamp(getattr(binding, 'workspace_created_at', None))
+        for binding in bindings
+    }
     resources = {
         resource['workspace_uuid']: resource for resource in await _cloud_workspace_resource_counts(ap, bindings)
     }
@@ -229,6 +248,7 @@ async def build_heartbeat_payloads(ap: core_app.Application) -> list[dict]:
         await build_heartbeat_payload(
             ap,
             workspace_uuid=workspace_uuid,
+            workspace_create_ts=workspace_create_ts.get(workspace_uuid, 0),
             workspace_resource=resources.get(workspace_uuid),
         )
         for workspace_uuid in workspace_uuids

src/langbot/pkg/telemetry/identity.py

@@ -4,13 +4,19 @@ import typing
 
 
 class WorkspaceExecutionContext(typing.Protocol):
+    @property
+    def instance_uuid(self) -> str: ...
+
     @property
     def workspace_uuid(self) -> str: ...
 
 
 def workspace_identity(execution_context: WorkspaceExecutionContext) -> dict[str, str]:
-    """Build the canonical telemetry identity for one Workspace execution."""
+    """Build both first-class telemetry identities for one execution."""
+    instance_id = execution_context.instance_uuid.strip()
     workspace_uuid = execution_context.workspace_uuid.strip()
+    if not instance_id:
+        raise ValueError('Telemetry execution instance ID is empty')
     if not workspace_uuid:
         raise ValueError('Telemetry execution Workspace UUID is empty')
-    return {'workspace_uuid': workspace_uuid}
+    return {'instance_id': instance_id, 'workspace_uuid': workspace_uuid}

src/langbot/pkg/telemetry/telemetry.py

@@ -136,12 +136,31 @@ class TelemetryManager:
                     try:
                         # Use asyncio.wait_for to ensure we always bound the total time
                         telemetry_token = os.getenv('LANGBOT_TELEMETRY_INGEST_TOKEN', '').strip()
+                        headers: dict[str, str] = {}
                         if telemetry_token:
-                            request = client.post(
-                                url,
-                                json=sanitized,
-                                headers={'X-LangBot-Telemetry-Token': telemetry_token},
-                            )
+                            headers['X-LangBot-Telemetry-Token'] = telemetry_token
+                        else:
+                            workspace_uuid = str(sanitized.get('workspace_uuid', '')).strip()
+                            user_service = getattr(self.ap, 'user_service', None)
+                            if workspace_uuid and user_service is not None:
+                                try:
+                                    owner = await user_service.get_workspace_owner(workspace_uuid)
+                                    owner_email = str(getattr(owner, 'user', '') or '').strip()
+                                    space_service = getattr(self.ap, 'space_service', None)
+                                    access_token = (
+                                        await space_service.get_valid_access_token(owner_email)
+                                        if owner_email and space_service is not None
+                                        else None
+                                    )
+                                    access_token = str(access_token or '').strip()
+                                    if access_token:
+                                        headers['Authorization'] = f'Bearer {access_token}'
+                                except Exception:
+                                    self.ap.logger.debug(
+                                        'Could not resolve authenticated telemetry reporter', exc_info=True
+                                    )
+                        if headers:
+                            request = client.post(url, json=sanitized, headers=headers)
                         else:
                             request = client.post(url, json=sanitized)
                         resp = await asyncio.wait_for(request, timeout=10 + 1)

src/langbot/pkg/vector/mgr.py

@@ -67,7 +67,7 @@ class VectorDBManager:
                 use_business_database = pgvector_config.get('use_business_database', False)
                 allowed_dimensions = pgvector_config.get(
                     'allowed_dimensions',
-                    [384, 512, 768, 1024, 1536],
+                    [384, 512, 768, 1024, 1536, 3072],
                 )
                 common_options = {
                     'use_business_database': use_business_database,

src/langbot/pkg/vector/vdbs/pgvector_db.py

@@ -6,7 +6,7 @@ from collections.abc import AsyncIterator
 from typing import Any
 
 import sqlalchemy
-from pgvector.sqlalchemy import Vector
+from pgvector.sqlalchemy import HALFVEC, Vector
 from sqlalchemy.dialects.postgresql import insert as postgresql_insert
 from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
 from sqlalchemy.orm import declarative_base
@@ -18,7 +18,7 @@ from langbot.pkg.vector.vdb import VectorDatabase
 
 Base = declarative_base()
 
-DEFAULT_ALLOWED_DIMENSIONS = (384, 512, 768, 1024, 1536)
+DEFAULT_ALLOWED_DIMENSIONS = (384, 512, 768, 1024, 1536, 3072)
 
 # pgvector schema only stores these metadata fields.
 _PG_SUPPORTED_FIELDS = {'text', 'file_id', 'chunk_uuid'}
@@ -321,7 +321,12 @@ class PgVectorDatabase(VectorDatabase):
         if len(query_embedding) != scope.embedding_dimension:
             raise ValueError(f'Query embedding must have the selected dimension {scope.embedding_dimension}')
 
-        typed_embedding = sqlalchemy.cast(PgVectorEntry.embedding, Vector(scope.embedding_dimension))
+        typed_embedding = sqlalchemy.cast(
+            PgVectorEntry.embedding,
+            HALFVEC(scope.embedding_dimension)
+            if scope.embedding_dimension > 2000
+            else Vector(scope.embedding_dimension),
+        )
         distance = typed_embedding.cosine_distance(query_embedding)
         statement = (
             sqlalchemy.select(

src/langbot/pkg/workspace/collaboration.py

@@ -17,6 +17,7 @@ from ..entity.persistence.user import AccountStatus, User
 from ..entity.persistence.workspace import (
     InvitationStatus,
     MembershipRole,
+    MembershipSource,
     MembershipStatus,
     Workspace,
     WorkspaceInvitation,
@@ -88,6 +89,7 @@ class ResolvedWorkspaceAccess:
 @dataclasses.dataclass(frozen=True, slots=True)
 class WorkspaceMemberView:
     membership: WorkspaceMembership
+    display_name: str
     email: str
 
 
@@ -294,7 +296,7 @@ class WorkspaceCollaborationService:
         async def operation(active_session: AsyncSession) -> list[WorkspaceMemberView]:
             await self._load_actor(active_session, workspace_uuid, actor)
             statement = (
-                sqlalchemy.select(WorkspaceMembership, User.user)
+                sqlalchemy.select(WorkspaceMembership, User.user, User.normalized_email)
                 .join(User, User.uuid == WorkspaceMembership.account_uuid)
                 .where(
                     WorkspaceMembership.workspace_uuid == workspace_uuid,
@@ -304,8 +306,12 @@ class WorkspaceCollaborationService:
                 .order_by(WorkspaceMembership.created_at, WorkspaceMembership.uuid)
             )
             return [
-                WorkspaceMemberView(membership=membership, email=email)
-                for membership, email in (await active_session.execute(statement)).all()
+                WorkspaceMemberView(
+                    membership=membership,
+                    display_name=display_name,
+                    email=email,
+                )
+                for membership, display_name, email in (await active_session.execute(statement)).all()
             ]
 
         return await self._run(operation, session=session, read_only=True)
@@ -478,6 +484,7 @@ class WorkspaceCollaborationService:
                     account_uuid=account_uuid,
                     role=invitation.role,
                     status=MembershipStatus.ACTIVE.value,
+                    source=MembershipSource.LOCAL.value,
                     invited_by_account_uuid=invitation.created_by_account_uuid,
                     joined_at=now,
                     projection_revision=0,
@@ -486,6 +493,7 @@ class WorkspaceCollaborationService:
             elif membership.status != MembershipStatus.ACTIVE.value:
                 membership.role = invitation.role
                 membership.status = MembershipStatus.ACTIVE.value
+                membership.source = MembershipSource.LOCAL.value
                 membership.invited_by_account_uuid = invitation.created_by_account_uuid
                 membership.joined_at = now
 
@@ -606,6 +614,8 @@ class WorkspaceCollaborationService:
     ) -> WorkspaceMembership:
         if role not in {item.value for item in MembershipRole}:
             raise MembershipPermissionError('Unknown Workspace role')
+        if role == MembershipRole.OWNER.value:
+            raise MembershipPermissionError('Workspace ownership cannot be transferred')
 
         async def operation(active_session: AsyncSession) -> WorkspaceMembership:
             await self._require_active_workspace(active_session, workspace_uuid)
@@ -617,8 +627,8 @@ class WorkspaceCollaborationService:
                 target_account_uuid,
             )
             self._require_can_manage_target(persisted_actor, target, new_role=role)
-            if target.role == MembershipRole.OWNER.value and role != MembershipRole.OWNER.value:
-                await self._require_another_owner(active_session, workspace_uuid, target.account_uuid)
+            if target.role == MembershipRole.OWNER.value:
+                raise LastOwnerError('The Workspace owner cannot be removed or demoted')
             target.role = role
             await active_session.flush()
             return target
@@ -644,7 +654,7 @@ class WorkspaceCollaborationService:
             )
             self._require_can_manage_target(persisted_actor, target)
             if target.role == MembershipRole.OWNER.value:
-                await self._require_another_owner(active_session, workspace_uuid, target.account_uuid)
+                raise LastOwnerError('The Workspace owner cannot be removed or demoted')
             target.status = MembershipStatus.REMOVED.value
             await active_session.flush()
             return target
@@ -751,26 +761,6 @@ class WorkspaceCollaborationService:
             raise WorkspaceNotFoundError('Workspace not found')
         return persisted_actor
 
-    async def _require_another_owner(
-        self,
-        session: AsyncSession,
-        workspace_uuid: str,
-        excluded_account_uuid: str,
-    ) -> None:
-        owners = (
-            await session.scalars(
-                sqlalchemy.select(WorkspaceMembership)
-                .where(
-                    WorkspaceMembership.workspace_uuid == workspace_uuid,
-                    WorkspaceMembership.status == MembershipStatus.ACTIVE.value,
-                    WorkspaceMembership.role == MembershipRole.OWNER.value,
-                )
-                .with_for_update()
-            )
-        ).all()
-        if not any(owner.account_uuid != excluded_account_uuid for owner in owners):
-            raise LastOwnerError('The last Workspace owner cannot be removed or demoted')
-
     def _require_actor_workspace(self, actor: WorkspaceMembership, workspace_uuid: str) -> None:
         if actor.workspace_uuid != workspace_uuid or actor.status != MembershipStatus.ACTIVE.value:
             raise WorkspaceNotFoundError('Workspace not found')

src/langbot/pkg/workspace/entities.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import datetime
 from dataclasses import dataclass
 
 
@@ -12,3 +13,4 @@ class WorkspaceExecutionBinding:
     placement_generation: int
     write_fenced: bool
     state: str
+    workspace_created_at: datetime.datetime | None = None

src/langbot/pkg/workspace/service.py

@@ -11,6 +11,7 @@ from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
 
 from ..entity.persistence.workspace import (
     MembershipRole,
+    MembershipSource,
     MembershipStatus,
     Workspace,
     WorkspaceExecutionSource,
@@ -283,6 +284,7 @@ class WorkspaceService:
                 placement_generation=execution_state.active_generation,
                 write_fenced=execution_state.write_fenced,
                 state=execution_state.state,
+                workspace_created_at=workspace.created_at,
             )
 
         binding = await self._run(operation, session=session)
@@ -450,6 +452,7 @@ class WorkspaceService:
                 account_uuid=account_uuid,
                 role=MembershipRole.OWNER.value,
                 status=MembershipStatus.ACTIVE.value,
+                source=MembershipSource.LOCAL.value,
                 joined_at=joined_at,
                 projection_revision=0,
             )
@@ -457,6 +460,7 @@ class WorkspaceService:
         else:
             membership.role = MembershipRole.OWNER.value
             membership.status = MembershipStatus.ACTIVE.value
+            membership.source = MembershipSource.LOCAL.value
             membership.joined_at = membership.joined_at or joined_at
 
         if workspace.created_by_account_uuid is None:

src/langbot/templates/config.yaml

@@ -201,7 +201,7 @@ vdb:
         # keep this false when deliberately using an external pgvector DB.
         use_business_database: false
         # Release migrations create one partial ANN index per enabled value.
-        allowed_dimensions: [384, 512, 768, 1024, 1536]
+        allowed_dimensions: [384, 512, 768, 1024, 1536, 3072]
         host: '127.0.0.1'
         port: 5433
         database: 'langbot'
@@ -245,6 +245,8 @@ storage:
         max_concurrency: 16
 plugin:
     enable: true
+    # Maximum time for the Runtime transport, handshake, and desired-state replay.
+    connect_timeout_seconds: 180.0
     runtime_ws_url: 'ws://langbot_plugin_runtime:5400/control/ws'
     enable_marketplace: true
     display_plugin_debug_url: 'ws://localhost:5401/plugin/debug/ws'

src/langbot/templates/embed/widget.js

@@ -1240,6 +1240,14 @@
     // Root container
     var root = document.createElement("div");
     root.id = "langbot-widget-root";
+    root.langbotDestroy = function () {
+      wsDisconnect();
+      if (state.historyReloadTimer) {
+        clearTimeout(state.historyReloadTimer);
+        state.historyReloadTimer = null;
+      }
+      root.remove();
+    };
     document.body.appendChild(root);
 
     var shadow = root.attachShadow({ mode: "open" });

src/langbot/templates/metadata/pipeline/ai.yaml

@@ -92,6 +92,7 @@ stages:
         default:
           primary: ''
           fallbacks: []
+          reasoning: {}
       - name: max-round
         label:
           en_US: Max Round

tests/integration/api/test_plugins_security.py

@@ -106,7 +106,12 @@ async def plugin_security_api(plugin_module):
     application.plugin_connector.require_workspace_context = AsyncMock()
     application.plugin_connector.list_plugins = AsyncMock(return_value=[raw_plugin])
     application.plugin_connector.get_plugin_info = AsyncMock(return_value=raw_plugin)
-    application.plugin_connector.get_debug_info = AsyncMock(return_value={'plugin_debug_key': 'runtime-debug-secret'})
+    application.plugin_connector.get_debug_info = AsyncMock(
+        return_value={
+            'plugin_debug_key': 'runtime-debug-secret',
+            'expires_at': '2026-08-04T12:00:00Z',
+        }
+    )
     application.plugin_connector.get_plugin_logs = AsyncMock(return_value=['private runtime line'])
     application.plugin_connector.set_plugin_config = AsyncMock()
 
@@ -232,8 +237,9 @@ async def test_debug_key_requires_resource_manage_permission(plugin_security_api
     assert (await allowed.get_json())['data'] == {
         'debug_url': 'http://localhost:5401',
         'plugin_debug_key': 'runtime-debug-secret',
+        'expires_at': '2026-08-04T12:00:00Z',
     }
-    application.plugin_connector.get_debug_info.assert_awaited_once_with()
+    application.plugin_connector.get_debug_info.assert_awaited_once()
 
 
 @pytest.mark.asyncio

tests/integration/api/test_smoke.py

@@ -9,6 +9,8 @@ Run: uv run pytest tests/integration/api/test_smoke.py -q
 
 from __future__ import annotations
 
+from types import SimpleNamespace
+
 import pytest
 from unittest.mock import MagicMock, AsyncMock, Mock
 
@@ -304,12 +306,34 @@ class TestUserInitEndpoint:
         data = await response.get_json()
         assert data['data'] == {
             'initialized': True,
+            'authenticated_invitation_acceptance_enabled': False,
             'password_login_enabled': True,
             'space_login_enabled': False,
         }
         fake_api_app.user_service.get_login_capabilities.assert_awaited_once_with()
         fake_api_app.user_service.get_first_user.assert_not_awaited()
 
+    @pytest.mark.asyncio
+    async def test_account_info_enables_authenticated_invitation_acceptance_in_cloud(
+        self, quart_test_client, fake_api_app
+    ):
+        fake_api_app.deployment = SimpleNamespace(mode='cloud')
+        fake_api_app.user_service.is_initialized.return_value = True
+        fake_api_app.user_service.get_login_capabilities = AsyncMock(
+            return_value={'password_login_enabled': True, 'space_login_enabled': True}
+        )
+
+        response = await quart_test_client.get('/api/v1/user/account-info')
+
+        assert response.status_code == 200
+        data = await response.get_json()
+        assert data['data'] == {
+            'initialized': True,
+            'authenticated_invitation_acceptance_enabled': True,
+            'password_login_enabled': False,
+            'space_login_enabled': True,
+        }
+
     @pytest.mark.asyncio
     async def test_recovery_key_resets_any_existing_account(self, quart_test_client, fake_api_app, monkeypatch):
         fake_api_app.user_service.is_initialized.return_value = True

tests/integration/api/test_support_admin_launch.py

@@ -333,7 +333,6 @@ async def test_support_admin_request_context_has_actor_owner_and_no_membership(s
     assert Permission.RESOURCE_MANAGE.value in permissions
     assert not permissions.intersection(
         {
-            Permission.OWNER_TRANSFER.value,
             Permission.MEMBER_VIEW.value,
             Permission.MEMBER_INVITE.value,
             Permission.MEMBER_UPDATE_ROLE.value,

tests/integration/api/test_user_space_oauth.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import datetime
 from types import SimpleNamespace
 from unittest.mock import AsyncMock, Mock
 from urllib.parse import parse_qs, urlsplit
@@ -14,6 +15,7 @@ from langbot.pkg.api.http.controller.groups.user import UserRouterGroup
 
 pytestmark = pytest.mark.integration
 WORKSPACE_UUID = '11111111-1111-4111-8111-111111111111'
+WORKSPACE_CREATED_AT = datetime.datetime(2026, 1, 2, 3, 4, 5, tzinfo=datetime.UTC)
 
 
 @pytest.fixture
@@ -58,6 +60,12 @@ async def space_oauth_api():
         return_value={'account_uuid': 'account-a', 'workspace_uuid': WORKSPACE_UUID}
     )
     application.workspace_collaboration_service.resolve_account_workspace = AsyncMock(return_value=access)
+    application.workspace_service.get_execution_binding = AsyncMock(
+        return_value=SimpleNamespace(
+            workspace_uuid=WORKSPACE_UUID,
+            workspace_created_at=WORKSPACE_CREATED_AT,
+        )
+    )
     application.space_service.get_oauth_authorize_url = Mock(
         side_effect=lambda redirect_uri, state: f'https://space.example/authorize?state={state}'
     )
@@ -234,7 +242,11 @@ async def test_login_callback_requires_and_consumes_server_state(space_oauth_api
     assert response.status_code == 200
     assert (await response.get_json())['data']['token'] == 'space-login-token'
     application.user_service.consume_space_oauth_state_details.assert_awaited_once_with('opaque-login-state', 'login')
-    application.space_service.exchange_oauth_code.assert_awaited_once_with('oauth-code')
+    application.space_service.exchange_oauth_code.assert_awaited_once_with(
+        'oauth-code',
+        [WORKSPACE_UUID],
+        {WORKSPACE_UUID: int(WORKSPACE_CREATED_AT.timestamp())},
+    )
 
 
 @pytest.mark.asyncio
@@ -272,9 +284,10 @@ async def test_space_credits_are_resolved_from_workspace_owner(space_oauth_api):
         '/api/v1/user/space-credits',
         headers={'Authorization': 'Bearer account-token', 'X-Workspace-Id': WORKSPACE_UUID},
     )
+    payload = await response.get_json()
 
     assert response.status_code == 200
-    assert (await response.get_json())['data'] == {
+    assert payload['data'] == {
         'credits': 25000,
         'owner_space_bound': True,
         'is_workspace_owner': True,
@@ -282,6 +295,31 @@ async def test_space_credits_are_resolved_from_workspace_owner(space_oauth_api):
     application.space_service.get_credits.assert_awaited_once_with('owner@example.com')
 
 
+@pytest.mark.asyncio
+async def test_cloud_workspace_owner_is_always_space_bound_after_login(space_oauth_api):
+    application, client = space_oauth_api
+    application.deployment.mode = 'cloud'
+    application.user_service.get_workspace_owner = AsyncMock(return_value=None)
+    application.space_service.get_credits = AsyncMock()
+    application.cloud_model_catalog_service = SimpleNamespace(
+        get_workspace_credits=lambda workspace_uuid: 25000 if workspace_uuid == WORKSPACE_UUID else None
+    )
+
+    response = await client.get(
+        '/api/v1/user/space-credits',
+        headers={'Authorization': 'Bearer account-token', 'X-Workspace-Id': WORKSPACE_UUID},
+    )
+    payload = await response.get_json()
+
+    assert response.status_code == 200
+    assert payload['data'] == {
+        'credits': 25000,
+        'owner_space_bound': True,
+        'is_workspace_owner': True,
+    }
+    application.space_service.get_credits.assert_not_awaited()
+
+
 @pytest.mark.asyncio
 async def test_bind_callback_uses_opaque_state_and_never_treats_it_as_jwt(space_oauth_api):
     application, client = space_oauth_api

tests/integration/api/test_workspaces.py

@@ -188,6 +188,7 @@ async def test_owner_invites_second_account_and_secret_is_not_persisted(workspac
     workspace_uuid = current['workspace']['uuid']
     assert current['membership']['role'] == 'owner'
     assert 'member.invite' in current['permissions']
+    assert 'owner.transfer' not in current['permissions']
 
     invite_response = await client.post(
         f'/api/v1/workspaces/{workspace_uuid}/invitations',
@@ -263,6 +264,14 @@ async def test_owner_invites_second_account_and_secret_is_not_persisted(workspac
     assert member_current['membership']['role'] == 'viewer'
     assert 'member.invite' not in member_current['permissions']
 
+    transfer_response = await client.patch(
+        f'/api/v1/workspaces/{workspace_uuid}/members/{member_current["membership"]["account_uuid"]}',
+        headers=_auth(owner_token, workspace_uuid),
+        json={'role': 'owner'},
+    )
+    assert transfer_response.status_code == 403
+    assert (await transfer_response.get_json())['code'] == 'permission_denied'
+
     forbidden_invite = await client.post(
         f'/api/v1/workspaces/{workspace_uuid}/invitations',
         headers=_auth(member_token, workspace_uuid),
@@ -272,6 +281,31 @@ async def test_owner_invites_second_account_and_secret_is_not_persisted(workspac
     assert (await forbidden_invite.get_json())['code'] == 'permission_denied'
 
 
+async def test_workspace_member_list_returns_display_name_and_email(workspace_api):
+    _, client, engine, owner_token = workspace_api
+
+    current_response = await client.get('/api/v1/workspaces/current', headers=_auth(owner_token))
+    current = (await current_response.get_json())['data']
+    workspace_uuid = current['workspace']['uuid']
+    owner_uuid = current['membership']['account_uuid']
+
+    async with engine.begin() as connection:
+        await connection.execute(
+            sqlalchemy.update(User).where(User.uuid == owner_uuid).values(user='Owner Display Name')
+        )
+
+    response = await client.get(
+        f'/api/v1/workspaces/{workspace_uuid}/members',
+        headers=_auth(owner_token, workspace_uuid),
+    )
+
+    assert response.status_code == 200
+    members = (await response.get_json())['data']['members']
+    assert len(members) == 1
+    assert members[0]['display_name'] == 'Owner Display Name'
+    assert members[0]['email'] == 'owner@example.com'
+
+
 async def test_oss_invitation_accept_requires_logout_before_registration(workspace_api):
     _, client, _, owner_token = workspace_api
 

tests/integration/persistence/test_membership_source_migration.py

@@ -0,0 +1,70 @@
+from __future__ import annotations
+
+import pytest
+import sqlalchemy as sa
+from sqlalchemy.ext.asyncio import create_async_engine
+
+from langbot.pkg.persistence.alembic_runner import run_alembic_stamp, run_alembic_upgrade
+
+
+@pytest.mark.asyncio
+async def test_membership_source_migration_backfills_existing_rows_as_local_and_enforces_constraint(tmp_path):
+    engine = create_async_engine(f'sqlite+aiosqlite:///{tmp_path / "membership-source.db"}')
+    try:
+        async with engine.begin() as connection:
+            await connection.execute(
+                sa.text(
+                    """
+                    CREATE TABLE workspace_memberships (
+                        uuid VARCHAR(36) PRIMARY KEY,
+                        workspace_uuid VARCHAR(36) NOT NULL,
+                        account_uuid VARCHAR(36) NOT NULL,
+                        role VARCHAR(32) NOT NULL,
+                        status VARCHAR(32) NOT NULL,
+                        projection_revision BIGINT NOT NULL DEFAULT 0,
+                        created_at DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+                        updated_at DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP
+                    )
+                    """
+                )
+            )
+            await connection.execute(
+                sa.text(
+                    """
+                    INSERT INTO workspace_memberships
+                        (uuid, workspace_uuid, account_uuid, role, status, projection_revision)
+                    VALUES
+                        ('00000000-0000-4000-8000-000000000001', 'workspace', 'local-account',
+                         'viewer', 'active', 0),
+                        ('00000000-0000-4000-8000-000000000002', 'workspace', 'cloud-account',
+                         'viewer', 'active', 0)
+                    """
+                )
+            )
+
+        await run_alembic_stamp(engine, '0019_single_workspace_owner')
+        await run_alembic_upgrade(engine, 'head')
+
+        async with engine.connect() as connection:
+            rows = (
+                await connection.execute(sa.text('SELECT uuid, source FROM workspace_memberships ORDER BY uuid'))
+            ).all()
+            columns = await connection.run_sync(
+                lambda sync_connection: {
+                    column['name']: column
+                    for column in sa.inspect(sync_connection).get_columns('workspace_memberships')
+                }
+            )
+        assert rows == [
+            ('00000000-0000-4000-8000-000000000001', 'local'),
+            ('00000000-0000-4000-8000-000000000002', 'local'),
+        ]
+        assert columns['source']['nullable'] is False
+
+        with pytest.raises(sa.exc.IntegrityError):
+            async with engine.begin() as connection:
+                await connection.execute(
+                    sa.text("UPDATE workspace_memberships SET source = 'guessed-from-user-source'")
+                )
+    finally:
+        await engine.dispose()

tests/integration/persistence/test_migrations.py

@@ -9,8 +9,11 @@ Run: uv run pytest tests/integration/persistence/test_migrations.py -q
 
 from __future__ import annotations
 
+import json
+
 import pytest
 import sqlalchemy
+from sqlalchemy import text
 from sqlalchemy.ext.asyncio import create_async_engine
 
 from langbot.pkg.entity.persistence.base import Base
@@ -95,6 +98,29 @@ class TestSQLiteMigrationBaseline:
 class TestSQLiteMigrationUpgrade:
     """Tests for upgrade to head workflow."""
 
+    @pytest.mark.asyncio
+    async def test_upgrade_from_published_space_launch_head_to_merged_head(self, sqlite_engine):
+        """A database released at the production-only 0016 head must remain upgradable."""
+        async with sqlite_engine.begin() as conn:
+            await conn.run_sync(Base.metadata.create_all)
+
+        await run_alembic_stamp(sqlite_engine, '0016_space_launch_replay')
+        await run_alembic_upgrade(sqlite_engine, 'head')
+
+        assert await get_alembic_current(sqlite_engine) == _get_script_head()
+        assert _get_script_head() == '0021_merge_reasoning_config'
+
+    @pytest.mark.asyncio
+    async def test_upgrade_from_reasoning_config_head_to_merged_head(self, sqlite_engine):
+        """A database that already ran the feature migration must remain upgradable."""
+        async with sqlite_engine.begin() as conn:
+            await conn.run_sync(Base.metadata.create_all)
+
+        await run_alembic_stamp(sqlite_engine, '0018_llm_reasoning_config')
+        await run_alembic_upgrade(sqlite_engine, 'head')
+
+        assert await get_alembic_current(sqlite_engine) == '0021_merge_reasoning_config'
+
     @pytest.mark.asyncio
     async def test_upgrade_from_baseline_to_head(self, sqlite_engine):
         """
@@ -190,6 +216,66 @@ class TestSQLiteMigrationUpgrade:
         await run_alembic_upgrade(sqlite_engine, 'head')
         assert await get_alembic_current(sqlite_engine) == _get_script_head()
 
+    @pytest.mark.asyncio
+    async def test_reasoning_config_migrates_existing_models(self, sqlite_engine):
+        """Upgrade from 0017 backfills reasoning config and keeps a database default."""
+        async with sqlite_engine.begin() as conn:
+            await conn.execute(
+                text(
+                    """
+                    CREATE TABLE llm_models (
+                        uuid VARCHAR(255) PRIMARY KEY,
+                        name VARCHAR(255) NOT NULL,
+                        provider_uuid VARCHAR(255) NOT NULL,
+                        abilities JSON NOT NULL,
+                        context_length INTEGER,
+                        extra_args JSON NOT NULL,
+                        prefered_ranking INTEGER NOT NULL DEFAULT 0
+                    )
+                    """
+                )
+            )
+            await conn.execute(
+                text(
+                    """
+                    INSERT INTO llm_models (
+                        uuid, name, provider_uuid, abilities, extra_args, prefered_ranking
+                    ) VALUES (
+                        'existing-model', 'Existing Model', 'provider', '[]', '{}', 0
+                    )
+                    """
+                )
+            )
+
+        await run_alembic_stamp(sqlite_engine, '0017_oss_workspace_identity')
+        await run_alembic_upgrade(sqlite_engine, 'head')
+
+        async with sqlite_engine.begin() as conn:
+            columns = await conn.run_sync(lambda sync_conn: sqlalchemy.inspect(sync_conn).get_columns('llm_models'))
+            reasoning_column = next(column for column in columns if column['name'] == 'reasoning_config')
+            assert reasoning_column['nullable'] is False
+
+            existing_value = (
+                await conn.execute(text("SELECT reasoning_config FROM llm_models WHERE uuid = 'existing-model'"))
+            ).scalar_one()
+            assert json.loads(existing_value) == {'level': 'provider_default'}
+
+            await conn.execute(
+                text(
+                    """
+                    INSERT INTO llm_models (
+                        uuid, name, provider_uuid, abilities, extra_args, prefered_ranking
+                    ) VALUES (
+                        'new-model', 'New Model', 'provider', '[]', '{}', 0
+                    )
+                    """
+                )
+            )
+            new_value = (
+                await conn.execute(text("SELECT reasoning_config FROM llm_models WHERE uuid = 'new-model'"))
+            ).scalar_one()
+            assert json.loads(new_value) == {'level': 'provider_default'}
+
 
 class TestSQLiteMigrationFreshDatabase:
     """Tests for fresh database workflow."""

tests/integration/persistence/test_pgvector_postgres.py

@@ -85,6 +85,32 @@ async def clean_database(postgres_engine: AsyncEngine):
     await clean()
 
 
+async def test_upgrade_adds_3072_dimension_index_and_constraint(
+    postgres_engine: AsyncEngine,
+    clean_database,
+) -> None:
+    async with postgres_engine.begin() as conn:
+        await conn.execute(text('CREATE EXTENSION IF NOT EXISTS vector'))
+        await conn.run_sync(Base.metadata.create_all)
+    await run_alembic_stamp(postgres_engine, '0010_scope_resources')
+    await run_alembic_upgrade(postgres_engine, 'head')
+
+    async with postgres_engine.connect() as conn:
+        constraint = await conn.scalar(
+            text(
+                'SELECT pg_get_constraintdef(oid) FROM pg_constraint '
+                "WHERE conrelid = 'langbot_vectors'::regclass "
+                "AND conname = 'ck_langbot_vectors_embedding_dimension_enabled'"
+            )
+        )
+        assert '3072' in constraint
+        index_definition = await conn.scalar(
+            text("SELECT indexdef FROM pg_indexes WHERE indexname = 'ix_langbot_vectors_hnsw_cosine_3072'")
+        )
+        assert 'halfvec(3072)' in index_definition
+        assert 'halfvec_cosine_ops' in index_definition
+
+
 async def test_legacy_upgrade_temporarily_suspends_and_restores_source_rls_for_unprivileged_owner(
     postgres_url: str,
     postgres_engine: AsyncEngine,

tests/integration/persistence/test_release_migration_postgres.py

@@ -92,7 +92,7 @@ def _application(postgres_url: str, *, runtime_role: str = 'langbot_runtime_not_
                     'use': 'pgvector',
                     'pgvector': {
                         'use_business_database': True,
-                        'allowed_dimensions': [384, 512, 768, 1024, 1536],
+                        'allowed_dimensions': [384, 512, 768, 1024, 1536, 3072],
                     },
                 },
             }

tests/integration/persistence/test_single_owner_migration.py

@@ -0,0 +1,100 @@
+from __future__ import annotations
+
+import pytest
+import sqlalchemy as sa
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+
+from langbot.pkg.entity.persistence.base import Base
+from langbot.pkg.entity.persistence.user import User
+from langbot.pkg.entity.persistence.workspace import Workspace, WorkspaceMembership
+from langbot.pkg.persistence.alembic_runner import run_alembic_stamp, run_alembic_upgrade
+
+
+@pytest.mark.asyncio
+async def test_single_owner_migration_demotes_historical_extra_owner_and_installs_unique_index(tmp_path):
+    engine = create_async_engine(f'sqlite+aiosqlite:///{tmp_path / "single-owner.db"}')
+    try:
+        async with engine.begin() as connection:
+            await connection.run_sync(Base.metadata.create_all)
+            await connection.execute(sa.text('DROP INDEX uq_workspace_memberships_one_active_owner'))
+
+        session_factory = async_sessionmaker(engine, expire_on_commit=False)
+        workspace_uuid = '00000000-0000-4000-8000-000000000001'
+        creator_uuid = '00000000-0000-4000-8000-000000000010'
+        promoted_uuid = '00000000-0000-4000-8000-000000000020'
+        async with session_factory() as session:
+            session.add_all(
+                [
+                    User(
+                        uuid=creator_uuid,
+                        user='creator@example.test',
+                        normalized_email='creator@example.test',
+                        password='hash',
+                        account_type='local',
+                    ),
+                    User(
+                        uuid=promoted_uuid,
+                        user='promoted@example.test',
+                        normalized_email='promoted@example.test',
+                        password='hash',
+                        account_type='local',
+                    ),
+                    Workspace(
+                        uuid=workspace_uuid,
+                        instance_uuid='instance-test',
+                        name='Workspace',
+                        slug='workspace',
+                        type='team',
+                        status='active',
+                        source='local',
+                        created_by_account_uuid=creator_uuid,
+                    ),
+                    WorkspaceMembership(
+                        uuid='00000000-0000-4000-8000-000000000100',
+                        workspace_uuid=workspace_uuid,
+                        account_uuid=creator_uuid,
+                        role='owner',
+                        status='active',
+                    ),
+                    WorkspaceMembership(
+                        uuid='00000000-0000-4000-8000-000000000200',
+                        workspace_uuid=workspace_uuid,
+                        account_uuid=promoted_uuid,
+                        role='owner',
+                        status='active',
+                    ),
+                ]
+            )
+            await session.commit()
+
+        await run_alembic_stamp(engine, '0018_merge_launch_replay')
+        await run_alembic_upgrade(engine, 'head')
+
+        async with engine.connect() as connection:
+            roles = dict(
+                (
+                    await connection.execute(
+                        sa.text(
+                            'SELECT account_uuid, role FROM workspace_memberships '
+                            'WHERE workspace_uuid = :workspace_uuid ORDER BY account_uuid'
+                        ),
+                        {'workspace_uuid': workspace_uuid},
+                    )
+                ).all()
+            )
+            assert roles == {creator_uuid: 'owner', promoted_uuid: 'admin'}
+            indexes = await connection.run_sync(
+                lambda sync_connection: {
+                    index['name'] for index in sa.inspect(sync_connection).get_indexes('workspace_memberships')
+                }
+            )
+            assert 'uq_workspace_memberships_one_active_owner' in indexes
+
+        with pytest.raises(sa.exc.IntegrityError):
+            async with engine.begin() as connection:
+                await connection.execute(
+                    sa.text("UPDATE workspace_memberships SET role = 'owner' WHERE account_uuid = :account_uuid"),
+                    {'account_uuid': promoted_uuid},
+                )
+    finally:
+        await engine.dispose()

tests/unit_tests/api/http/test_authz.py

@@ -27,10 +27,9 @@ def test_owner_has_every_fixed_permission():
     assert ctx.workspace.permissions == frozenset(permission.value for permission in authz.Permission)
 
 
-def test_admin_cannot_transfer_owner_delete_workspace_or_link_billing():
+def test_admin_cannot_delete_workspace_or_link_billing():
     ctx = _context(authz.WorkspaceRole.ADMIN)
 
-    assert not authz.has_permission(ctx, authz.Permission.OWNER_TRANSFER)
     assert not authz.has_permission(ctx, authz.Permission.WORKSPACE_DELETE)
     assert not authz.has_permission(ctx, authz.Permission.BILLING_LINK_MANAGE)
     assert authz.has_permission(ctx, authz.Permission.MEMBER_INVITE)

tests/unit_tests/api/service/test_bot_service.py

@@ -9,8 +9,9 @@ Source: src/langbot/pkg/api/http/service/bot.py
 from __future__ import annotations
 
 import pytest
-from unittest.mock import AsyncMock, Mock, patch
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
 from types import SimpleNamespace
+import json
 import uuid
 
 from langbot.pkg.api.http.service.bot import BotService
@@ -241,6 +242,29 @@ class TestBotServiceGetRuntimeBotInfo:
         assert result['adapter_runtime_values']['webhook_url'] == '/bots/wecom-uuid'
         assert result['adapter_runtime_values']['webhook_full_url'] == 'http://127.0.0.1:5300/bots/wecom-uuid'
 
+    async def test_get_runtime_bot_info_returns_webhook_for_http_bot(self):
+        ap = SimpleNamespace(
+            instance_config=SimpleNamespace(
+                data={'api': {'webhook_prefix': 'https://bot.example.com'}}
+            ),
+            platform_mgr=SimpleNamespace(get_bot_by_uuid=AsyncMock(return_value=None)),
+        )
+        service = BotService(ap)
+        service.get_bot = AsyncMock(
+            return_value={
+                'uuid': 'http-bot-uuid',
+                'name': 'HTTP Bot',
+                'adapter': 'http_bot',
+                'adapter_config': {},
+            }
+        )
+
+        result = await service.get_runtime_bot_info(WORKSPACE_UUID, 'http-bot-uuid')
+
+        assert result['adapter_runtime_values']['webhook_full_url'] == (
+            'https://bot.example.com/bots/http-bot-uuid'
+        )
+
     async def test_get_runtime_bot_info_no_webhook_for_telegram(self):
         """Returns no webhook URL for non-webhook adapters like telegram."""
         # Setup
@@ -605,6 +629,77 @@ class TestBotServiceListEventLogs:
         assert total == 5
 
 
+class TestBotServiceHttpBotInboundTest:
+    async def test_sends_signed_message_through_public_ingress(self):
+        ap = SimpleNamespace(
+            instance_config=SimpleNamespace(data={'api': {'port': 5300}}),
+        )
+        service = BotService(ap)
+        service.get_bot = AsyncMock(
+            return_value={
+                'uuid': 'http-bot-uuid',
+                'adapter': 'http_bot',
+                'adapter_config': {
+                    'signature_required': True,
+                    'inbound_secret': 'test-secret',
+                },
+                'enable': True,
+            }
+        )
+        response = MagicMock(status=202)
+        session = MagicMock()
+        session.post.return_value.__aenter__ = AsyncMock(return_value=response)
+        session.post.return_value.__aexit__ = AsyncMock(return_value=None)
+
+        with (
+            patch('langbot.pkg.api.http.service.bot.httpclient.get_session', return_value=session),
+            patch(
+                'langbot.pkg.api.http.service.bot.httpclient.read_json_limited',
+                new=AsyncMock(
+                    return_value={
+                        'code': 0,
+                        'data': {
+                            'session_id': 'wizard-session',
+                            'accepted_message_id': 'in-message',
+                        },
+                    }
+                ),
+            ),
+        ):
+            result = await service.send_http_bot_test_message(
+                WORKSPACE_UUID,
+                'http-bot-uuid',
+                'hello',
+            )
+
+        assert result['accepted_message_id'] == 'in-message'
+        request = session.post.call_args
+        assert request.args[0] == 'http://127.0.0.1:5300/bots/http-bot-uuid'
+        payload = json.loads(request.kwargs['data'])
+        assert payload['message'] == [{'type': 'Plain', 'text': 'hello'}]
+        headers = request.kwargs['headers']
+        assert headers['X-LB-Timestamp']
+        assert headers['X-LB-Signature'].startswith('sha256=')
+
+    async def test_rejects_non_http_bot(self):
+        service = BotService(SimpleNamespace())
+        service.get_bot = AsyncMock(
+            return_value={
+                'uuid': 'telegram-bot',
+                'adapter': 'telegram',
+                'adapter_config': {},
+                'enable': True,
+            }
+        )
+
+        with pytest.raises(ValueError, match='only available for HTTP Bot'):
+            await service.send_http_bot_test_message(
+                WORKSPACE_UUID,
+                'telegram-bot',
+                'hello',
+            )
+
+
 class TestBotServiceSendMessage:
     """Tests for send_message method."""
 

tests/unit_tests/api/service/test_cloud_managed_model_protection.py

@@ -0,0 +1,112 @@
+"""Cloud Runtime write protection for the managed LangBot Models catalog."""
+
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import pytest
+
+from langbot.pkg.api.http.service import model as model_service_module
+from langbot.pkg.api.http.service.model import (
+    EmbeddingModelsService,
+    LLMModelsService,
+    RerankModelsService,
+    _assert_cloud_managed_provider_mutable,
+)
+from langbot.pkg.cloud.model_catalog import LANGBOT_MODELS_PROVIDER_REQUESTER
+
+
+WORKSPACE = 'workspace-a'
+PROVIDER = 'managed-provider'
+MODEL = 'managed-model'
+
+
+@pytest.mark.asyncio
+async def test_managed_provider_guard_is_cloud_only(monkeypatch) -> None:
+    async def managed_provider(_ap, _context, provider_uuid):
+        assert provider_uuid == PROVIDER
+        return {'uuid': PROVIDER, 'requester': LANGBOT_MODELS_PROVIDER_REQUESTER}
+
+    monkeypatch.setattr(model_service_module, '_require_workspace_provider', managed_provider)
+    application = SimpleNamespace(persistence_mgr=SimpleNamespace(mode=SimpleNamespace(value='cloud_runtime')))
+
+    with pytest.raises(ValueError, match='managed by Cloud'):
+        await _assert_cloud_managed_provider_mutable(
+            application,
+            WORKSPACE,
+            PROVIDER,
+        )
+
+    application.persistence_mgr.mode.value = 'normal'
+    await _assert_cloud_managed_provider_mutable(
+        application,
+        WORKSPACE,
+        PROVIDER,
+    )
+
+
+@pytest.mark.parametrize(
+    ('service_type', 'create_method', 'model_data'),
+    [
+        (LLMModelsService, 'create_llm_model', {'provider_uuid': PROVIDER, 'name': 'chat', 'abilities': []}),
+        (EmbeddingModelsService, 'create_embedding_model', {'provider_uuid': PROVIDER, 'name': 'embedding'}),
+        (RerankModelsService, 'create_rerank_model', {'provider_uuid': PROVIDER, 'name': 'rerank'}),
+    ],
+)
+@pytest.mark.asyncio
+async def test_all_model_types_reject_creation_under_managed_provider(
+    monkeypatch,
+    service_type,
+    create_method: str,
+    model_data: dict,
+) -> None:
+    guard = AsyncMock(side_effect=ValueError('LangBot Models is managed by Cloud and cannot be modified'))
+    monkeypatch.setattr(model_service_module, '_assert_cloud_managed_provider_mutable', guard)
+    application = SimpleNamespace(
+        persistence_mgr=SimpleNamespace(),
+        provider_service=SimpleNamespace(
+            get_provider=AsyncMock(return_value={'uuid': PROVIDER, 'requester': LANGBOT_MODELS_PROVIDER_REQUESTER})
+        ),
+        model_mgr=None,
+    )
+    service = service_type(application)
+
+    with pytest.raises(ValueError, match='managed by Cloud'):
+        await getattr(service, create_method)(WORKSPACE, model_data)
+
+    guard.assert_awaited_once()
+
+
+@pytest.mark.parametrize(
+    ('service_type', 'get_method', 'write_method', 'payload'),
+    [
+        (LLMModelsService, 'get_llm_model', 'update_llm_model', {'name': 'changed'}),
+        (LLMModelsService, 'get_llm_model', 'delete_llm_model', None),
+        (EmbeddingModelsService, 'get_embedding_model', 'update_embedding_model', {'name': 'changed'}),
+        (EmbeddingModelsService, 'get_embedding_model', 'delete_embedding_model', None),
+        (RerankModelsService, 'get_rerank_model', 'update_rerank_model', {'name': 'changed'}),
+        (RerankModelsService, 'get_rerank_model', 'delete_rerank_model', None),
+    ],
+)
+@pytest.mark.asyncio
+async def test_all_model_types_reject_update_and_delete_for_managed_provider(
+    monkeypatch,
+    service_type,
+    get_method: str,
+    write_method: str,
+    payload: dict | None,
+) -> None:
+    guard = AsyncMock(side_effect=ValueError('LangBot Models is managed by Cloud and cannot be modified'))
+    monkeypatch.setattr(model_service_module, '_assert_cloud_managed_provider_mutable', guard)
+    application = SimpleNamespace(persistence_mgr=SimpleNamespace(mode=SimpleNamespace(value='cloud_runtime')))
+    service = service_type(application)
+    monkeypatch.setattr(
+        service,
+        get_method,
+        AsyncMock(return_value={'uuid': MODEL, 'provider_uuid': PROVIDER, 'extra_args': {}}),
+    )
+
+    args = (WORKSPACE, MODEL) if payload is None else (WORKSPACE, MODEL, payload)
+    with pytest.raises(ValueError, match='managed by Cloud'):
+        await getattr(service, write_method)(*args)
+
+    guard.assert_awaited_once()

tests/unit_tests/api/service/test_model_service.py

@@ -17,12 +17,14 @@ import pytest
 from unittest.mock import AsyncMock, Mock
 from types import SimpleNamespace
 
+from langbot.pkg.api.http.context import ExecutionContext
 from langbot.pkg.api.http.service.model import (
     LLMModelsService,
     EmbeddingModelsService,
     RerankModelsService,
     _parse_provider_api_keys,
     _runtime_model_data,
+    _serialize_llm_model,
     _validate_provider_supports,
 )
 from langbot.pkg.api.http.service import model as model_service_module
@@ -64,15 +66,19 @@ def _create_mock_llm_model(
     abilities: list = None,
     context_length: int | None = None,
     extra_args: dict = None,
+    reasoning_config: dict = None,
 ) -> Mock:
     """Helper to create mock LLMModel entity."""
     model = Mock(spec=LLMModel)
+    model.workspace_uuid = WORKSPACE_UUID
     model.uuid = model_uuid
     model.name = name
     model.provider_uuid = provider_uuid
     model.abilities = abilities or []
     model.context_length = context_length
     model.extra_args = extra_args or {}
+    model.reasoning_config = reasoning_config or {'level': 'provider_default'}
+    model.prefered_ranking = 0
     return model
 
 
@@ -156,6 +162,26 @@ def _create_runtime_model_mgr() -> SimpleNamespace:
     return manager
 
 
+def _create_reasoning_runtime_provider(capabilities: dict) -> SimpleNamespace:
+    execution_context = ExecutionContext(
+        instance_uuid='instance-test',
+        workspace_uuid=WORKSPACE_UUID,
+        placement_generation=1,
+    )
+    return SimpleNamespace(
+        execution_context=execution_context,
+        provider_entity=ModelProvider(
+            workspace_uuid=WORKSPACE_UUID,
+            uuid='provider-uuid',
+            name='Reasoning Provider',
+            requester='openai',
+            base_url='https://api.openai.com',
+            api_keys=[],
+        ),
+        requester=SimpleNamespace(get_reasoning_capabilities=Mock(return_value=capabilities)),
+    )
+
+
 class TestParseProviderApiKeys:
     """Tests for _parse_provider_api_keys helper function."""
 
@@ -209,6 +235,42 @@ class TestRuntimeModelData:
         assert result['extra_args'] == {'temp': 0.7}
 
 
+class TestSerializeLLMModel:
+    def test_includes_runtime_reasoning_capabilities(self):
+        model = _create_mock_llm_model(
+            abilities=['reasoning'],
+            reasoning_config={'level': 'high'},
+        )
+        capabilities = {
+            'supported': True,
+            'levels': ['provider_default', 'low', 'high'],
+            'source': 'litellm',
+        }
+        runtime_model = SimpleNamespace(
+            model_entity=model,
+            provider=SimpleNamespace(
+                requester=SimpleNamespace(get_reasoning_capabilities=Mock(return_value=capabilities))
+            ),
+        )
+        ap = SimpleNamespace(
+            persistence_mgr=SimpleNamespace(
+                serialize_model=Mock(
+                    return_value={
+                        'uuid': model.uuid,
+                        'name': model.name,
+                        'reasoning_config': {'level': 'high'},
+                    }
+                )
+            ),
+            model_mgr=SimpleNamespace(llm_model_dict={('workspace', model.uuid): runtime_model}),
+        )
+
+        serialized = _serialize_llm_model(ap, model)
+
+        assert serialized['reasoning_config'] == {'level': 'high'}
+        assert serialized['reasoning_capabilities'] == capabilities
+
+
 class TestLLMModelsServiceGetLLMModels:
     """Tests for LLMModelsService.get_llm_models method."""
 
@@ -580,6 +642,66 @@ class TestLLMModelsServiceCreateLLMModel:
         ap.provider_service.find_or_create_provider.assert_called_once()
         assert result_uuid is not None
 
+    async def test_create_llm_model_validates_explicit_reasoning_level(self):
+        ap = SimpleNamespace()
+        ap.persistence_mgr = SimpleNamespace(execute_async=AsyncMock(return_value=_create_mock_result([])))
+        runtime_provider = _create_reasoning_runtime_provider(
+            {
+                'supported': True,
+                'levels': ['provider_default', 'low', 'high'],
+                'source': 'litellm',
+            }
+        )
+        ap.model_mgr = _create_runtime_model_mgr()
+        ap.model_mgr.provider_dict = {'provider-uuid': runtime_provider}
+
+        service = LLMModelsService(ap)
+        await service.create_llm_model(
+            WORKSPACE_UUID,
+            {
+                'uuid': 'reasoning-model',
+                'name': 'Reasoning Model',
+                'provider_uuid': 'provider-uuid',
+                'abilities': ['reasoning'],
+                'reasoning_config': {'level': 'high'},
+                'extra_args': {},
+            },
+            preserve_uuid=True,
+            auto_set_to_default_pipeline=False,
+        )
+
+        runtime_entity = ap.model_mgr.load_llm_model_with_provider.await_args.args[1]
+        assert runtime_entity.reasoning_config == {'level': 'high'}
+
+    async def test_create_llm_model_rejects_unsupported_reasoning_before_insert(self):
+        ap = SimpleNamespace()
+        ap.persistence_mgr = SimpleNamespace(execute_async=AsyncMock())
+        runtime_provider = _create_reasoning_runtime_provider(
+            {
+                'supported': True,
+                'levels': ['provider_default'],
+                'source': 'manual',
+            }
+        )
+        ap.model_mgr = _create_runtime_model_mgr()
+        ap.model_mgr.provider_dict = {'provider-uuid': runtime_provider}
+
+        service = LLMModelsService(ap)
+        with pytest.raises(ValueError, match='Available levels: provider_default'):
+            await service.create_llm_model(
+                WORKSPACE_UUID,
+                {
+                    'name': 'Unknown Reasoning Model',
+                    'provider_uuid': 'provider-uuid',
+                    'abilities': ['reasoning'],
+                    'reasoning_config': {'level': 'high'},
+                    'extra_args': {},
+                },
+                auto_set_to_default_pipeline=False,
+            )
+
+        ap.persistence_mgr.execute_async.assert_not_awaited()
+
 
 class TestLLMModelsServiceUpdateLLMModel:
     """Tests for LLMModelsService.update_llm_model method."""
@@ -595,7 +717,10 @@ class TestLLMModelsServiceUpdateLLMModel:
         ap.model_mgr.remove_llm_model = AsyncMock()
         ap.model_mgr.load_llm_model_with_provider = AsyncMock(return_value=Mock())
 
-        ap.persistence_mgr.execute_async = AsyncMock()
+        existing_model = _create_mock_llm_model()
+        ap.persistence_mgr.execute_async = AsyncMock(
+            side_effect=[_create_mock_result(first_item=existing_model), _create_mock_result()]
+        )
 
         service = LLMModelsService(ap)
         service.get_llm_model = AsyncMock(return_value=_existing_llm_data())
@@ -623,7 +748,8 @@ class TestLLMModelsServiceUpdateLLMModel:
         ap.model_mgr.provider_dict = {}  # Empty
         ap.model_mgr.remove_llm_model = AsyncMock()
 
-        ap.persistence_mgr.execute_async = AsyncMock()
+        existing_model = _create_mock_llm_model()
+        ap.persistence_mgr.execute_async = AsyncMock(return_value=_create_mock_result(first_item=existing_model))
 
         service = LLMModelsService(ap)
         service.get_llm_model = AsyncMock(return_value=_existing_llm_data('nonexistent-provider'))

tests/unit_tests/api/service/test_provider_service.py

@@ -25,6 +25,7 @@ from langbot.pkg.workspace.errors import WorkspaceNotFoundError
 pytestmark = pytest.mark.asyncio
 
 WORKSPACE_UUID = 'workspace-a'
+SYSTEM_REQUESTER = 'space-chat-completions'
 
 
 def _create_mock_provider(
@@ -1005,3 +1006,56 @@ class TestProviderSecretRoundtrip:
             )
 
         ap.persistence_mgr.execute_async.assert_not_awaited()
+
+
+class TestCloudManagedProviderProtection:
+    @staticmethod
+    def _service() -> ModelProviderService:
+        ap = SimpleNamespace(
+            persistence_mgr=SimpleNamespace(
+                mode=SimpleNamespace(value='cloud_runtime'),
+                execute_async=AsyncMock(),
+            ),
+            model_mgr=SimpleNamespace(),
+        )
+        return ModelProviderService(ap)
+
+    async def test_cloud_rejects_user_created_system_requester(self):
+        service = self._service()
+
+        with pytest.raises(ValueError, match='reserved'):
+            await service.create_provider(
+                WORKSPACE_UUID,
+                {
+                    'name': 'Fake LangBot Models',
+                    'requester': SYSTEM_REQUESTER,
+                    'base_url': 'https://example.invalid/v1',
+                    'api_keys': ['fake'],
+                },
+            )
+
+        with pytest.raises(ValueError, match='reserved'):
+            await service.find_or_create_provider(
+                WORKSPACE_UUID,
+                SYSTEM_REQUESTER,
+                'https://api.langbot.cloud/v1',
+                ['fake'],
+            )
+        service.ap.persistence_mgr.execute_async.assert_not_awaited()
+
+    async def test_cloud_rejects_update_and_delete_of_managed_provider(self):
+        service = self._service()
+        service.get_provider = AsyncMock(
+            return_value={'uuid': 'system-provider', 'requester': SYSTEM_REQUESTER}
+        )
+
+        with pytest.raises(ValueError, match='managed by Cloud'):
+            await service.update_provider(WORKSPACE_UUID, 'system-provider', {'name': 'Renamed'})
+        with pytest.raises(ValueError, match='managed by Cloud'):
+            await service.delete_provider(WORKSPACE_UUID, 'system-provider')
+        service.ap.persistence_mgr.execute_async.assert_not_awaited()
+
+    async def test_oss_does_not_reserve_space_requester(self):
+        ap = SimpleNamespace(persistence_mgr=SimpleNamespace(mode=SimpleNamespace(value='oss_compat')))
+        service = ModelProviderService(ap)
+        assert service._system_requester_is_reserved(SYSTEM_REQUESTER) is False

tests/unit_tests/api/service/test_space_service.py

@@ -25,6 +25,7 @@ import time
 
 from langbot.pkg.api.http.service.space import SpaceService
 from langbot.pkg.entity.persistence.user import User
+from langbot.pkg.utils import constants
 
 
 pytestmark = pytest.mark.asyncio
@@ -573,10 +574,20 @@ class TestSpaceServiceExchangeOAuthCode:
             mock_session_obj.post.return_value.__aexit__ = AsyncMock(return_value=None)
 
             # Execute
-            result = await service.exchange_oauth_code('auth_code')
+            result = await service.exchange_oauth_code(
+                'auth_code',
+                ['workspace-1'],
+                {'workspace-1': 1_700_000_000},
+            )
 
         # Verify
         assert result['access_token'] == 'new_access_token'
+        assert mock_session_obj.post.call_args.kwargs['json'] == {
+            'code': 'auth_code',
+            'instance_id': constants.instance_id,
+            'workspace_uuids': ['workspace-1'],
+            'workspace_created_ats': {'workspace-1': 1_700_000_000},
+        }
 
     async def test_exchange_oauth_code_api_error(self):
         """Raises ValueError on API error."""
@@ -809,6 +820,100 @@ class TestSpaceServiceGetModels:
                 await service.get_models()
 
 
+class TestSpaceServiceGetModelSelection:
+    """Tests for availability-ranked model selection."""
+
+    @pytest.mark.parametrize('response_shape', ['direct', 'models-envelope', 'availability-wrapper'])
+    async def test_preserves_selection_order_and_category_query(self, response_shape):
+        ap = SimpleNamespace(instance_config=SimpleNamespace(data={}))
+        service = SpaceService(ap)
+        models = [
+            {
+                'uuid': 'best-model',
+                'model_id': 'best-chat-model',
+                'provider': 'provider-1',
+                'category': 'chat',
+                'status': 'active',
+            },
+            {
+                'uuid': 'fallback-model',
+                'model_id': 'fallback-chat-model',
+                'provider': 'provider-2',
+                'category': 'chat',
+                'status': 'active',
+            },
+        ]
+        if response_shape == 'models-envelope':
+            data = {'models': models}
+        elif response_shape == 'availability-wrapper':
+            data = [
+                {'model': model, 'latency_ms': index + 10, 'http_code': 200}
+                for index, model in enumerate(models)
+            ]
+        else:
+            data = models
+        payload = {'code': 0, 'data': data}
+        mock_response = MagicMock(status=200)
+
+        with (
+            patch('langbot.pkg.api.http.service.space.httpclient.get_session') as get_session,
+            patch(
+                'langbot.pkg.api.http.service.space.httpclient.read_json_limited',
+                new=AsyncMock(return_value=payload),
+            ),
+        ):
+            session = MagicMock()
+            session.get.return_value.__aenter__ = AsyncMock(return_value=mock_response)
+            session.get.return_value.__aexit__ = AsyncMock(return_value=None)
+            get_session.return_value = session
+
+            result = await service.get_model_selection('chat')
+
+        assert [model.uuid for model in result] == ['best-model', 'fallback-model']
+        session.get.assert_called_once_with(
+            'https://space.langbot.app/api/v1/models/selection',
+            params={'category': 'chat'},
+        )
+
+    async def test_recommended_model_uses_first_selection_and_refreshes_once(self):
+        local_model = SimpleNamespace(uuid='local-model-uuid', name='best-chat-model')
+        persistence = SimpleNamespace(
+            execute_async=AsyncMock(
+                side_effect=[
+                    _create_mock_result(first_item=None),
+                    _create_mock_result(first_item=local_model),
+                ]
+            )
+        )
+        model_mgr = SimpleNamespace(sync_new_models_from_space=AsyncMock())
+        ap = SimpleNamespace(
+            instance_config=SimpleNamespace(data={}),
+            persistence_mgr=persistence,
+            model_mgr=model_mgr,
+        )
+        service = SpaceService(ap)
+        service.get_model_selection = AsyncMock(
+            return_value=[
+                SimpleNamespace(uuid='best-upstream-uuid', model_id='best-chat-model'),
+                SimpleNamespace(uuid='fallback-upstream-uuid', model_id='fallback-chat-model'),
+            ]
+        )
+        context = SimpleNamespace(
+            instance_uuid='instance',
+            workspace_uuid='workspace',
+            placement_generation=1,
+            principal=SimpleNamespace(),
+            entitlement_revision=0,
+        )
+
+        result = await service.get_recommended_chat_model(context)
+
+        assert result == {'uuid': 'local-model-uuid', 'name': 'best-chat-model'}
+        service.get_model_selection.assert_awaited_once_with('chat')
+        model_mgr.sync_new_models_from_space.assert_awaited_once()
+        assert persistence.execute_async.await_count == 2
+
+
 class TestSpaceServiceCreditsCache:
     """Tests for credits cache behavior."""
 

tests/unit_tests/api/test_plugin_runtime_route_fence.py

@@ -124,24 +124,37 @@ async def test_background_plugin_operation_refences_captured_generation(plugin_r
 
 
 @pytest.mark.asyncio
-async def test_background_plugin_operation_revalidates_inside_short_tenant_uow(plugin_router_cls):
+async def test_background_plugin_operation_revalidates_and_runs_inside_tenant_uow(plugin_router_cls):
     scopes = []
+    active_scope = None
+
+    transaction_active = False
 
     @asynccontextmanager
-    async def tenant_uow(workspace_uuid):
+    async def tenant_scope(workspace_uuid):
+        nonlocal active_scope
         scopes.append(workspace_uuid)
-        yield
+        active_scope = workspace_uuid
+        try:
+            yield
+        finally:
+            active_scope = None
 
     connector = SimpleNamespace(
         require_workspace_context=AsyncMock(side_effect=lambda context: context),
     )
-    operation = AsyncMock(return_value='done')
+
+    async def operation():
+        assert active_scope == CONTEXT.workspace_uuid
+        assert transaction_active is False
+        return 'done'
+
     router = object.__new__(plugin_router_cls)
     router.ap = SimpleNamespace(
         plugin_connector=connector,
         persistence_mgr=SimpleNamespace(
             mode=SimpleNamespace(value='cloud_runtime'),
-            tenant_uow=tenant_uow,
+            tenant_scope=tenant_scope,
         ),
     )
 
@@ -150,4 +163,3 @@ async def test_background_plugin_operation_revalidates_inside_short_tenant_uow(p
     assert result == 'done'
     assert scopes == [CONTEXT.workspace_uuid]
     connector.require_workspace_context.assert_awaited_once_with(CONTEXT)
-    operation.assert_awaited_once()

tests/unit_tests/cloud/test_bootstrap.py

@@ -66,6 +66,10 @@ class _Provider:
     def __init__(self):
         self.manifest_provider = _Manifest()
 
+    async def fetch_model_catalog(self, instance_uuid: str):
+        del instance_uuid
+        raise AssertionError('not used by bootstrap contract tests')
+
     def bootstrap(self, *, instance_uuid: str, instance_config: dict):
         del instance_config
         return VerifiedCloudDeployment(
@@ -79,6 +83,7 @@ class _Provider:
             entitlement_provider=_Entitlements(),
             directory_provider=_Directory(),
             manifest_provider=self.manifest_provider,
+            model_catalog_provider=self,
             verification_key_id='root-2026',
         )
 
@@ -103,7 +108,7 @@ def _cloud_config() -> dict:
             'use': 'pgvector',
             'pgvector': {
                 'use_business_database': True,
-                'allowed_dimensions': [384, 768, 1536],
+                'allowed_dimensions': [384, 768, 1536, 3072],
             },
         },
         'mcp': {'stdio': {'enabled': False}},
@@ -211,7 +216,6 @@ async def test_cloud_directory_capacity_contract_is_fail_closed(directory_config
     [
         ({'use_business_database': False, 'allowed_dimensions': [1536]}, 'use_business_database=true'),
         ({'use_business_database': True, 'allowed_dimensions': []}, 'allowed_dimensions'),
-        ({'use_business_database': True, 'allowed_dimensions': [3072]}, 'allowed_dimensions'),
         ({'use_business_database': True, 'allowed_dimensions': [True]}, 'allowed_dimensions'),
     ],
 )
@@ -228,10 +232,23 @@ async def test_cloud_pgvector_contract_is_fail_closed(pgvector_config, message):
         )
 
 
+async def test_cloud_runtime_allows_explicitly_disabled_box():
+    config = _cloud_config()
+    config['box']['enabled'] = False
+
+    deployment = await resolve_deployment(
+        instance_uuid='instance-a',
+        instance_config=config,
+        entry_points=lambda: _EntryPoints([_EntryPoint(_Provider())]),
+        now=1_000,
+    )
+
+    assert isinstance(deployment, VerifiedCloudDeployment)
+
+
 @pytest.mark.parametrize(
     ('mutate', 'message'),
     [
-        (lambda config: config['box'].update(enabled=False), 'box.enabled=true'),
         (lambda config: config['box'].update(backend='docker'), 'box.backend=nsjail'),
         (lambda config: config['box']['runtime'].update(endpoint=''), 'box.runtime.endpoint'),
         (

tests/unit_tests/cloud/test_directory_projection.py

@@ -181,6 +181,39 @@ def _delta(
     )
 
 
+async def test_directory_delta_requests_model_catalog_sync_after_commit(projection_context):
+    application, _session_factory = projection_context
+    request_sync = Mock()
+    application.cloud_model_catalog_service = SimpleNamespace(request_sync=request_sync)
+    event = DirectoryEvent(
+        cursor=2,
+        uuid='20000000-0000-4000-8000-000000000002',
+        aggregate_uuid=WORKSPACE_UUID,
+        event_type='directory.changed',
+        revision=2,
+        payload={'workspace_uuid': WORKSPACE_UUID, 'directory_revision': 2},
+        created_at=datetime.datetime(2026, 7, 24, 12, 30, tzinfo=datetime.UTC),
+    )
+    batch = DirectoryEventBatch(
+        instance_uuid=INSTANCE_UUID,
+        after_cursor=1,
+        cursor=2,
+        high_water_cursor=2,
+        events=[event],
+    )
+    service = DirectoryProjectionService(
+        application,
+        _Provider([_snapshot(1)], [batch], [_delta(workspaces=[_workspace(revision=2)])]),
+        INSTANCE_UUID,
+    )
+    await service.initialize()
+    request_sync.reset_mock()
+
+    await service.sync_once()
+
+    request_sync.assert_called_once_with()
+
+
 async def test_initial_snapshot_projects_core_owned_rows(projection_context):
     application, session_factory = projection_context
     reconcile_execution_projection = Mock()
@@ -1023,7 +1056,7 @@ async def test_snapshot_for_another_instance_is_rejected(projection_context):
         await service.initialize()
 
 
-async def test_core_owned_membership_survives_directory_updates_and_omission(projection_context):
+async def test_directory_revision_zero_membership_is_adopted(projection_context):
     application, session_factory = projection_context
     service = DirectoryProjectionService(application, _Provider([_snapshot(1)]), INSTANCE_UUID)
     await service.initialize()
@@ -1034,29 +1067,125 @@ async def test_core_owned_membership_survives_directory_updates_and_omission(pro
             membership.role = 'viewer'
             membership.status = 'active'
             membership.projection_revision = 0
-            session.add(
-                WorkspaceMembership(
-                    uuid=SECOND_MEMBERSHIP_UUID,
-                    workspace_uuid=WORKSPACE_UUID,
-                    account_uuid='20000000-0000-0000-0000-000000000099',
-                    role='viewer',
-                    status='active',
-                    joined_at=membership.joined_at,
-                    projection_revision=0,
-                )
-            )
 
     projected_member = _member(revision=2).model_copy(update={'role': 'owner', 'membership_status': 'removed'})
     projected_workspace = _workspace(revision=2).model_copy(update={'members': (projected_member,)})
     await service.apply_snapshot(_snapshot(2, workspaces=[projected_workspace]))
 
     async with session_factory() as session:
-        memberships = {
-            membership.uuid: membership
-            for membership in (await session.scalars(sqlalchemy.select(WorkspaceMembership))).all()
-        }
-    assert memberships[MEMBERSHIP_UUID].role == 'viewer'
-    assert memberships[MEMBERSHIP_UUID].status == 'active'
-    assert memberships[MEMBERSHIP_UUID].projection_revision == 0
-    assert memberships[SECOND_MEMBERSHIP_UUID].status == 'active'
-    assert memberships[SECOND_MEMBERSHIP_UUID].projection_revision == 0
+        membership = await session.scalar(sqlalchemy.select(WorkspaceMembership))
+    assert membership.source == 'cloud_projection'
+    assert membership.role == 'owner'
+    assert membership.status == 'removed'
+    assert membership.projection_revision == 2
+
+
+async def test_directory_revision_zero_membership_omitted_from_snapshot_is_removed(projection_context):
+    application, session_factory = projection_context
+    service = DirectoryProjectionService(application, _Provider([_snapshot(1)]), INSTANCE_UUID)
+    await service.initialize()
+
+    historical_account_uuid = '20000000-0000-0000-0000-000000000099'
+    async with session_factory() as session:
+        async with session.begin():
+            membership = await session.scalar(sqlalchemy.select(WorkspaceMembership))
+            session.add(
+                User(
+                    uuid=historical_account_uuid,
+                    user='Historical Space Member',
+                    normalized_email='historical@example.com',
+                    password='',
+                    status='active',
+                    source='cloud_projection',
+                    projection_revision=1,
+                    account_type='space',
+                    space_account_uuid=historical_account_uuid,
+                )
+            )
+            session.add(
+                WorkspaceMembership(
+                    uuid=SECOND_MEMBERSHIP_UUID,
+                    workspace_uuid=WORKSPACE_UUID,
+                    account_uuid=historical_account_uuid,
+                    role='viewer',
+                    status='active',
+                    source='cloud_projection',
+                    joined_at=membership.joined_at,
+                    projection_revision=0,
+                )
+            )
+
+    await service.apply_snapshot(_snapshot(2))
+
+    async with session_factory() as session:
+        historical = await session.get(WorkspaceMembership, SECOND_MEMBERSHIP_UUID)
+    assert historical.status == 'removed'
+    assert historical.projection_revision == 2
+
+
+async def test_cloud_account_core_invitation_membership_survives_directory_omission(projection_context):
+    application, session_factory = projection_context
+    service = DirectoryProjectionService(application, _Provider([_snapshot(1)]), INSTANCE_UUID)
+    await service.initialize()
+
+    invited_account_uuid = '20000000-0000-0000-0000-000000000098'
+    async with session_factory() as session:
+        async with session.begin():
+            projected_membership = await session.scalar(sqlalchemy.select(WorkspaceMembership))
+            session.add(
+                User(
+                    uuid=invited_account_uuid,
+                    user='Invited Cloud Account',
+                    normalized_email='invited-cloud@example.com',
+                    password='',
+                    status='active',
+                    source='cloud_projection',
+                    projection_revision=1,
+                    account_type='space',
+                    space_account_uuid=invited_account_uuid,
+                )
+            )
+            session.add(
+                WorkspaceMembership(
+                    uuid=SECOND_MEMBERSHIP_UUID,
+                    workspace_uuid=WORKSPACE_UUID,
+                    account_uuid=invited_account_uuid,
+                    role='viewer',
+                    status='active',
+                    source='local',
+                    joined_at=projected_membership.joined_at,
+                    projection_revision=0,
+                )
+            )
+
+    await service.apply_snapshot(_snapshot(2))
+
+    async with session_factory() as session:
+        membership = await session.get(WorkspaceMembership, SECOND_MEMBERSHIP_UUID)
+    assert membership.source == 'local'
+    assert membership.status == 'active'
+    assert membership.projection_revision == 0
+
+
+async def test_directory_does_not_adopt_local_membership_with_different_uuid_for_same_cloud_account(projection_context):
+    application, session_factory = projection_context
+    service = DirectoryProjectionService(application, _Provider([_snapshot(1)]), INSTANCE_UUID)
+    await service.initialize()
+
+    async with session_factory() as session:
+        async with session.begin():
+            membership = await session.scalar(sqlalchemy.select(WorkspaceMembership))
+            membership.uuid = SECOND_MEMBERSHIP_UUID
+            membership.source = 'local'
+            membership.projection_revision = 0
+
+    projected_member = _member(revision=2).model_copy(update={'role': 'owner', 'membership_status': 'removed'})
+    projected_workspace = _workspace(revision=2).model_copy(update={'members': (projected_member,)})
+    await service.apply_snapshot(_snapshot(2, workspaces=[projected_workspace]))
+
+    async with session_factory() as session:
+        membership = await session.get(WorkspaceMembership, SECOND_MEMBERSHIP_UUID)
+    assert membership.source == 'local'
+    assert membership.role == 'developer'
+    assert membership.status == 'active'
+    assert membership.projection_revision == 0

tests/unit_tests/cloud/test_model_catalog.py

@@ -0,0 +1,466 @@
+from __future__ import annotations
+
+import asyncio
+import logging
+from datetime import UTC, datetime
+from types import SimpleNamespace
+
+import pytest
+import sqlalchemy
+from sqlalchemy.ext.asyncio import create_async_engine
+
+from langbot.pkg.cloud.model_catalog import (
+    CloudModelCatalogSnapshot,
+    CloudModelCatalogSyncService,
+    system_model_uuid,
+    system_provider_uuid,
+)
+from langbot.pkg.entity.persistence.base import Base
+from langbot.pkg.entity.persistence.model import EmbeddingModel, LLMModel, ModelProvider
+from langbot.pkg.entity.persistence.workspace import Workspace
+from langbot.pkg.persistence.mgr import PersistenceManager, PersistenceMode
+
+
+pytestmark = pytest.mark.asyncio
+INSTANCE_UUID = 'instance-model-catalog'
+WORKSPACE_A = '00000000-0000-4000-8000-000000000001'
+WORKSPACE_B = '00000000-0000-4000-8000-000000000002'
+OWNER_A = '10000000-0000-4000-8000-000000000001'
+OWNER_B = '10000000-0000-4000-8000-000000000002'
+
+
+class _CatalogProvider:
+    def __init__(self, snapshot: CloudModelCatalogSnapshot) -> None:
+        self.snapshot = snapshot
+
+    async def fetch_model_catalog(self, instance_uuid: str) -> CloudModelCatalogSnapshot:
+        assert instance_uuid == INSTANCE_UUID
+        return self.snapshot
+
+
+def _snapshot(
+    *,
+    key_a: str | None = 'owner-a-key',
+    model_id: str = 'gpt-test',
+    include_embedding: bool = True,
+) -> CloudModelCatalogSnapshot:
+    models = [
+        {
+            'uuid': 'upstream-chat',
+            'model_id': model_id,
+            'category': 'chat',
+            'llm_abilities': ['chat', 'vision'],
+            'is_featured': True,
+            'featured_order': 7,
+        }
+    ]
+    if include_embedding:
+        models.append(
+            {
+                'uuid': 'upstream-embedding',
+                'model_id': 'embedding-test',
+                'category': 'embedding',
+            }
+        )
+    return CloudModelCatalogSnapshot.model_validate(
+        {
+            'instance_uuid': INSTANCE_UUID,
+            'generated_at': datetime.now(UTC),
+            'base_url': 'https://api.langbot.cloud/v1/',
+            'models': models,
+            'workspaces': [
+                {
+                    'workspace_uuid': WORKSPACE_A,
+                    'owner_account_uuid': OWNER_A,
+                    'api_key': key_a,
+                    'credits': 25000,
+                },
+                {
+                    'workspace_uuid': WORKSPACE_B,
+                    'owner_account_uuid': OWNER_B,
+                    'api_key': 'owner-b-key',
+                    'credits': 5000,
+                },
+            ],
+        }
+    )
+
+
+async def test_catalog_snapshot_treats_null_model_abilities_as_empty() -> None:
+    payload = _snapshot().model_dump(mode='json')
+    payload['models'][0]['llm_abilities'] = None
+
+    snapshot = CloudModelCatalogSnapshot.model_validate(payload)
+
+    assert snapshot.models[0].llm_abilities == ()
+
+
+async def test_catalog_reconciles_every_workspace_idempotently_and_tracks_owner_and_downlisting(tmp_path) -> None:
+    engine = create_async_engine(f'sqlite+aiosqlite:///{tmp_path / "model-catalog.db"}')
+    manager = PersistenceManager(object(), mode=PersistenceMode.CLOUD_RUNTIME)
+    manager.db = SimpleNamespace(get_engine=lambda: engine)
+    bindings = [
+        SimpleNamespace(instance_uuid=INSTANCE_UUID, workspace_uuid=WORKSPACE_A, placement_generation=1),
+        SimpleNamespace(instance_uuid=INSTANCE_UUID, workspace_uuid=WORKSPACE_B, placement_generation=1),
+    ]
+    workspace_service = SimpleNamespace(list_active_execution_bindings=lambda: _async_value(bindings))
+    reload_counter = _AsyncCounter()
+    runtime_reload = SimpleNamespace(load_models_from_db=reload_counter)
+    app = SimpleNamespace(
+        persistence_mgr=manager,
+        workspace_service=workspace_service,
+        model_mgr=runtime_reload,
+        logger=logging.getLogger(__name__),
+    )
+    provider = _CatalogProvider(_snapshot())
+    service = CloudModelCatalogSyncService(app, provider, INSTANCE_UUID)
+
+    try:
+        async with engine.begin() as connection:
+            await connection.run_sync(Base.metadata.create_all)
+            await connection.execute(
+                sqlalchemy.insert(Workspace),
+                [
+                    {
+                        'uuid': WORKSPACE_A,
+                        'instance_uuid': INSTANCE_UUID,
+                        'name': 'A',
+                        'slug': 'a',
+                        'source': 'cloud_projection',
+                    },
+                    {
+                        'uuid': WORKSPACE_B,
+                        'instance_uuid': INSTANCE_UUID,
+                        'name': 'B',
+                        'slug': 'b',
+                        'source': 'cloud_projection',
+                    },
+                ],
+            )
+            await connection.execute(
+                sqlalchemy.insert(ModelProvider).values(
+                    uuid='custom-provider',
+                    workspace_uuid=WORKSPACE_A,
+                    name='Custom',
+                    requester='openai-chat-completions',
+                    base_url='https://custom.example/v1',
+                    api_keys=['custom-key'],
+                )
+            )
+            await connection.execute(
+                sqlalchemy.insert(LLMModel).values(
+                    uuid='custom-model',
+                    workspace_uuid=WORKSPACE_A,
+                    name='custom-model',
+                    provider_uuid='custom-provider',
+                    abilities=['chat'],
+                    extra_args={},
+                    prefered_ranking=0,
+                )
+            )
+
+        first = await service.sync_once()
+        assert first == {'workspaces': 2, 'created': 6, 'updated': 0, 'deleted': 0}
+        assert reload_counter.calls == 1
+        assert service.get_workspace_credits(WORKSPACE_A) == 25000
+        assert service.get_workspace_credits(WORKSPACE_B) == 5000
+
+        async with engine.connect() as connection:
+            providers = (
+                await connection.execute(
+                    sqlalchemy.select(
+                        ModelProvider.uuid,
+                        ModelProvider.workspace_uuid,
+                        ModelProvider.api_keys,
+                    ).where(ModelProvider.requester == 'space-chat-completions')
+                )
+            ).all()
+            assert {item.workspace_uuid for item in providers} == {WORKSPACE_A, WORKSPACE_B}
+            assert {item.uuid for item in providers} == {
+                system_provider_uuid(WORKSPACE_A),
+                system_provider_uuid(WORKSPACE_B),
+            }
+            assert {item.workspace_uuid: item.api_keys for item in providers} == {
+                WORKSPACE_A: ['owner-a-key'],
+                WORKSPACE_B: ['owner-b-key'],
+            }
+            assert await connection.scalar(sqlalchemy.select(sqlalchemy.func.count()).select_from(LLMModel)) == 3
+            assert await connection.scalar(sqlalchemy.select(sqlalchemy.func.count()).select_from(EmbeddingModel)) == 2
+
+        second = await service.sync_once()
+        assert second == {'workspaces': 2, 'created': 0, 'updated': 0, 'deleted': 0}
+        assert reload_counter.calls == 1
+
+        provider.snapshot = _snapshot(
+            key_a='new-owner-key',
+            model_id='gpt-renamed',
+            include_embedding=False,
+        )
+        third = await service.sync_once()
+        assert third == {'workspaces': 2, 'created': 0, 'updated': 3, 'deleted': 2}
+        assert reload_counter.calls == 2
+
+        async with engine.connect() as connection:
+            provider_a_keys = await connection.scalar(
+                sqlalchemy.select(ModelProvider.api_keys).where(ModelProvider.uuid == system_provider_uuid(WORKSPACE_A))
+            )
+            assert provider_a_keys == ['new-owner-key']
+            system_model_names = (
+                (
+                    await connection.execute(
+                        sqlalchemy.select(LLMModel.name).where(
+                            LLMModel.provider_uuid.in_(
+                                [system_provider_uuid(WORKSPACE_A), system_provider_uuid(WORKSPACE_B)]
+                            )
+                        )
+                    )
+                )
+                .scalars()
+                .all()
+            )
+            assert set(system_model_names) == {'gpt-renamed'}
+            assert await connection.scalar(sqlalchemy.select(sqlalchemy.func.count()).select_from(EmbeddingModel)) == 0
+            assert (
+                await connection.scalar(
+                    sqlalchemy.select(sqlalchemy.func.count())
+                    .select_from(ModelProvider)
+                    .where(ModelProvider.uuid == 'custom-provider')
+                )
+                == 1
+            )
+            assert (
+                await connection.scalar(
+                    sqlalchemy.select(sqlalchemy.func.count())
+                    .select_from(LLMModel)
+                    .where(LLMModel.uuid == 'custom-model')
+                )
+                == 1
+            )
+
+        provider.snapshot = _snapshot(key_a=None, model_id='gpt-renamed', include_embedding=False)
+        fourth = await service.sync_once()
+        assert fourth == {'workspaces': 2, 'created': 0, 'updated': 1, 'deleted': 0}
+        assert reload_counter.calls == 3
+        async with engine.connect() as connection:
+            provider_a_keys = await connection.scalar(
+                sqlalchemy.select(ModelProvider.api_keys).where(ModelProvider.uuid == system_provider_uuid(WORKSPACE_A))
+            )
+            assert provider_a_keys == []
+    finally:
+        await engine.dispose()
+
+
+def test_workspace_scoped_ids_are_stable_and_secrets_are_redacted() -> None:
+    assert system_provider_uuid(WORKSPACE_A) == system_provider_uuid(WORKSPACE_A)
+    assert system_provider_uuid(WORKSPACE_A) != system_provider_uuid(WORKSPACE_B)
+    assert system_model_uuid(WORKSPACE_A, 'chat', 'upstream') != system_model_uuid(WORKSPACE_B, 'chat', 'upstream')
+    snapshot = _snapshot()
+    assert 'owner-a-key' not in repr(snapshot)
+
+
+async def test_snapshot_must_cover_every_active_workspace() -> None:
+    snapshot = _snapshot().model_copy(update={'workspaces': _snapshot().workspaces[:1]})
+    app = SimpleNamespace(
+        workspace_service=SimpleNamespace(
+            list_active_execution_bindings=lambda: _async_value(
+                [SimpleNamespace(workspace_uuid=WORKSPACE_A), SimpleNamespace(workspace_uuid=WORKSPACE_B)]
+            )
+        ),
+        logger=logging.getLogger(__name__),
+    )
+    service = CloudModelCatalogSyncService(app, _CatalogProvider(snapshot), INSTANCE_UUID)
+    with pytest.raises(ValueError, match='missing billing projections for 1 active Workspaces'):
+        await service.sync_once()
+
+
+async def test_periodic_sync_discovers_workspace_created_after_startup_cache_release(tmp_path) -> None:
+    engine = create_async_engine(f'sqlite+aiosqlite:///{tmp_path / "model-catalog-new-workspace.db"}')
+    manager = PersistenceManager(object(), mode=PersistenceMode.CLOUD_RUNTIME)
+    manager.db = SimpleNamespace(get_engine=lambda: engine)
+    startup_bindings = [
+        SimpleNamespace(instance_uuid=INSTANCE_UUID, workspace_uuid=WORKSPACE_A, placement_generation=1)
+    ]
+    live_bindings = [
+        *startup_bindings,
+        SimpleNamespace(instance_uuid=INSTANCE_UUID, workspace_uuid=WORKSPACE_B, placement_generation=1),
+    ]
+
+    class _WorkspaceService:
+        startup_released = False
+
+        async def list_active_execution_bindings(self):
+            return list(live_bindings if self.startup_released else startup_bindings)
+
+        def release_startup_execution_bindings(self):
+            self.startup_released = True
+
+    workspace_service = _WorkspaceService()
+    app = SimpleNamespace(
+        persistence_mgr=manager,
+        workspace_service=workspace_service,
+        model_mgr=SimpleNamespace(load_models_from_db=_AsyncCounter()),
+        logger=logging.getLogger(__name__),
+    )
+    service = CloudModelCatalogSyncService(app, _CatalogProvider(_snapshot()), INSTANCE_UUID)
+
+    try:
+        async with engine.begin() as connection:
+            await connection.run_sync(Base.metadata.create_all)
+            await connection.execute(
+                sqlalchemy.insert(Workspace),
+                [
+                    {
+                        'uuid': WORKSPACE_A,
+                        'instance_uuid': INSTANCE_UUID,
+                        'name': 'A',
+                        'slug': 'a',
+                        'source': 'cloud_projection',
+                    },
+                    {
+                        'uuid': WORKSPACE_B,
+                        'instance_uuid': INSTANCE_UUID,
+                        'name': 'B',
+                        'slug': 'b',
+                        'source': 'cloud_projection',
+                    },
+                ],
+            )
+
+        await service.initialize()
+        workspace_service.release_startup_execution_bindings()
+        await service.sync_once()
+
+        async with engine.connect() as connection:
+            provider_b = await connection.scalar(
+                sqlalchemy.select(ModelProvider).where(ModelProvider.uuid == system_provider_uuid(WORKSPACE_B))
+            )
+        assert provider_b is not None
+    finally:
+        await engine.dispose()
+
+
+async def test_catalog_run_wakes_immediately_when_directory_changes() -> None:
+    sync_started = asyncio.Event()
+
+    class _WakeService(CloudModelCatalogSyncService):
+        async def sync_once(self, *, reload_runtime: bool = True):
+            del reload_runtime
+            sync_started.set()
+            return {'workspaces': 0, 'created': 0, 'updated': 0, 'deleted': 0}
+
+    app = SimpleNamespace(logger=logging.getLogger(__name__))
+    service = _WakeService(app, _CatalogProvider(_snapshot()), INSTANCE_UUID, sync_interval_seconds=3600)
+    task = asyncio.create_task(service.run())
+    try:
+        await asyncio.sleep(0)
+        service.request_sync()
+        await asyncio.wait_for(sync_started.wait(), timeout=0.2)
+    finally:
+        task.cancel()
+        with pytest.raises(asyncio.CancelledError):
+            await task
+
+
+async def _async_value(value):
+    return value
+
+
+class _AsyncCounter:
+    def __init__(self) -> None:
+        self.calls = 0
+
+    async def __call__(self) -> None:
+        self.calls += 1
+
+
+async def test_partial_workspace_failure_reloads_already_committed_changes() -> None:
+    bindings = [
+        SimpleNamespace(workspace_uuid=WORKSPACE_A),
+        SimpleNamespace(workspace_uuid=WORKSPACE_B),
+    ]
+    reload_counter = _AsyncCounter()
+    app = SimpleNamespace(
+        workspace_service=SimpleNamespace(list_active_execution_bindings=lambda: _async_value(bindings)),
+        model_mgr=SimpleNamespace(load_models_from_db=reload_counter),
+        logger=logging.getLogger(__name__),
+    )
+    service = CloudModelCatalogSyncService(app, _CatalogProvider(_snapshot()), INSTANCE_UUID)
+    calls = 0
+
+    async def sync_workspace(*_args):
+        nonlocal calls
+        calls += 1
+        if calls == 1:
+            return {'created': 1, 'updated': 0, 'deleted': 0}
+        raise RuntimeError('second Workspace failed')
+
+    service._sync_workspace = sync_workspace  # type: ignore[method-assign]
+
+    with pytest.raises(RuntimeError, match='second Workspace failed'):
+        await service.sync_once()
+    assert service.get_workspace_credits(WORKSPACE_A) == 25000
+    assert service.get_workspace_credits(WORKSPACE_B) is None
+    assert reload_counter.calls == 1
+
+
+async def test_failed_runtime_reload_is_retried_after_noop_sync() -> None:
+    bindings = [SimpleNamespace(workspace_uuid=WORKSPACE_A)]
+
+    class _FlakyReload:
+        def __init__(self) -> None:
+            self.calls = 0
+
+        async def __call__(self) -> None:
+            self.calls += 1
+            if self.calls == 1:
+                raise RuntimeError('reload failed')
+
+    runtime_reload = _FlakyReload()
+    app = SimpleNamespace(
+        workspace_service=SimpleNamespace(list_active_execution_bindings=lambda: _async_value(bindings)),
+        model_mgr=SimpleNamespace(load_models_from_db=runtime_reload),
+        logger=logging.getLogger(__name__),
+    )
+    service = CloudModelCatalogSyncService(app, _CatalogProvider(_snapshot()), INSTANCE_UUID)
+    calls = 0
+
+    async def sync_workspace(*_args):
+        nonlocal calls
+        calls += 1
+        if calls == 1:
+            return {'created': 1, 'updated': 0, 'deleted': 0}
+        return {'created': 0, 'updated': 0, 'deleted': 0}
+
+    service._sync_workspace = sync_workspace  # type: ignore[method-assign]
+
+    with pytest.raises(RuntimeError, match='reload failed'):
+        await service.sync_once()
+    summary = await service.sync_once()
+    assert summary == {'workspaces': 1, 'created': 0, 'updated': 0, 'deleted': 0}
+    assert runtime_reload.calls == 2
+
+
+async def test_background_sync_log_redacts_exception_message(caplog) -> None:
+    secret = 'owner-secret-api-key'
+    attempted = asyncio.Event()
+
+    class _FailingProvider:
+        async def fetch_model_catalog(self, instance_uuid: str) -> CloudModelCatalogSnapshot:
+            del instance_uuid
+            attempted.set()
+            raise RuntimeError(f'database parameters include {secret}')
+
+    app = SimpleNamespace(logger=logging.getLogger(__name__))
+    service = CloudModelCatalogSyncService(app, _FailingProvider(), INSTANCE_UUID)
+    service.sync_interval_seconds = 0.001
+    task = asyncio.create_task(service.run())
+    try:
+        await asyncio.wait_for(attempted.wait(), timeout=1)
+        await asyncio.sleep(0.01)
+    finally:
+        task.cancel()
+        with pytest.raises(asyncio.CancelledError):
+            await task
+
+    assert secret not in caplog.text
+    assert 'Cloud model catalog synchronization failed (RuntimeError)' in caplog.text

tests/unit_tests/core/test_load_config.py

@@ -319,6 +319,7 @@ class TestApplyEnvOverridesToConfig:
         load_config = get_load_config_module()
         cfg = {
             'plugin': {
+                'connect_timeout_seconds': 30.0,
                 'worker': {
                     'max_cpus': 1.0,
                     'max_memory_mb': 512,
@@ -329,11 +330,12 @@ class TestApplyEnvOverridesToConfig:
                     'restart_failure_threshold': 8,
                     'restart_failure_window_seconds': 30.0,
                     'restart_circuit_open_seconds': 60.0,
-                }
+                },
             },
             'mcp': {'stdio': {'enabled': True}},
         }
         env = {
+            'PLUGIN__CONNECT_TIMEOUT_SECONDS': '180',
             'PLUGIN__WORKER__MAX_CPUS': '2.5',
             'PLUGIN__WORKER__MAX_MEMORY_MB': '1024',
             'PLUGIN__WORKER__MAX_PIDS': '64',
@@ -349,6 +351,7 @@ class TestApplyEnvOverridesToConfig:
         with patch.dict(os.environ, env, clear=True):
             result = load_config._apply_env_overrides_to_config(cfg)
 
+        assert result['plugin']['connect_timeout_seconds'] == 180.0
         assert result['plugin']['worker'] == {
             'max_cpus': 2.5,
             'max_memory_mb': 1024,
@@ -393,6 +396,14 @@ class TestApplyEnvOverridesToConfig:
         assert isinstance(result['plugin']['worker']['max_memory_mb'], int)
         assert result['mcp']['stdio']['enabled'] is False
 
+    def test_runtime_policy_defaults_add_typed_plugin_connect_timeout(self):
+        load_config = get_load_config_module()
+
+        completed = load_config._complete_runtime_policy_defaults({'plugin': {'enable': True}})
+
+        assert completed['plugin']['connect_timeout_seconds'] == 180.0
+        assert isinstance(completed['plugin']['connect_timeout_seconds'], float)
+
     def test_webhook_prefix_override(self):
         """Test overriding webhook_prefix via environment variable."""
         load_config = get_load_config_module()

tests/unit_tests/persistence/test_tenant_uow.py

@@ -7,7 +7,7 @@ from types import SimpleNamespace
 
 import pytest
 import sqlalchemy as sa
-from pgvector.sqlalchemy import Vector
+from pgvector.sqlalchemy import HALFVEC, Vector
 from sqlalchemy.dialects.postgresql import insert as postgresql_insert
 from sqlalchemy.dialects.sqlite import insert as sqlite_insert
 from sqlalchemy.ext.asyncio import create_async_engine
@@ -961,11 +961,13 @@ async def test_scoped_session_rejects_raw_or_unapproved_sql(
         sa.select(sa.func.coalesce(sa.func.sum(sa.literal(1)), sa.literal(0))),
         sa.select(
             sa.func.now(),
+            sa.func.date_trunc('hour', sa.column('timestamp')),
             sa.func.length(sa.literal('value')),
             sa.func.nullif(sa.literal('value'), sa.literal('')),
         ),
         sa.select(sa.column('embedding').op('<=>')(sa.literal([0.1]))),
         sa.select(sa.cast(sa.column('embedding'), Vector(384))),
+        sa.select(sa.cast(sa.column('embedding'), HALFVEC(3072))),
         sa.insert(sa.table('rows', sa.column('id'))).values(id=1),
         _multi_value_statement(value=1),
         _on_conflict_statement(update_value=sa.func.coalesce(sa.literal(1), sa.literal(0))),

tests/unit_tests/pipeline/test_controller_tenancy.py

@@ -11,6 +11,7 @@ from sqlalchemy.ext.asyncio import create_async_engine
 from langbot.pkg.persistence.mgr import PersistenceManager, PersistenceMode
 from langbot.pkg.persistence.tenant_uow import PersistenceScopeKind
 from langbot.pkg.pipeline.controller import Controller
+from langbot.pkg.pipeline.pool import QueryPool
 from langbot.pkg.workspace.errors import WorkspaceGenerationMismatchError
 
 
@@ -28,6 +29,57 @@ def _prepare_scheduler(mock_app):
     return query_pool, session
 
 
+@pytest.mark.asyncio
+async def test_consumer_schedules_query_after_running_transition(
+    mock_app,
+    sample_query,
+):
+    query_pool = MagicMock()
+    query_pool.queries = [sample_query]
+    query_pool.__aenter__ = AsyncMock(return_value=query_pool)
+    query_pool.__aexit__ = AsyncMock(return_value=None)
+    query_pool.remove_query = AsyncMock(return_value=True)
+    wait_for_query = asyncio.Event()
+    query_pool.condition = SimpleNamespace(
+        wait=AsyncMock(side_effect=wait_for_query.wait),
+        notify_all=Mock(),
+    )
+    query_pool.mark_query_running_locked = Mock(side_effect=query_pool.queries.remove)
+    mock_app.query_pool = query_pool
+
+    session = SimpleNamespace(_semaphore=asyncio.Semaphore(1))
+    mock_app.sess_mgr.get_session = AsyncMock(return_value=session)
+    runtime_pipeline = SimpleNamespace(run=AsyncMock())
+    mock_app.pipeline_mgr = SimpleNamespace(get_pipeline_by_uuid=AsyncMock(return_value=runtime_pipeline))
+
+    task_created = asyncio.Event()
+    process_tasks = []
+
+    def create_process_task(coro, **_kwargs):
+        process_tasks.append(asyncio.create_task(coro))
+        task_created.set()
+
+    mock_app.task_mgr.create_task = Mock(side_effect=create_process_task)
+    controller = Controller(mock_app)
+    initial_slots = controller.semaphore._value
+    consumer_task = asyncio.create_task(controller.consumer())
+
+    try:
+        await asyncio.wait_for(task_created.wait(), timeout=2)
+    finally:
+        consumer_task.cancel()
+        with pytest.raises(asyncio.CancelledError):
+            await consumer_task
+        await asyncio.gather(*process_tasks)
+
+    query_pool.mark_query_running_locked.assert_called_once_with(sample_query)
+    runtime_pipeline.run.assert_awaited_once_with(sample_query)
+    query_pool.remove_query.assert_awaited_once_with(sample_query)
+    assert query_pool.queries == []
+    assert session._semaphore._value == 1
+    assert controller.semaphore._value == initial_slots
+
+
 @pytest.mark.asyncio
 async def test_controller_drops_stale_query_before_pipeline_lookup(
     mock_app,
@@ -143,3 +195,31 @@ async def test_controller_revalidates_generation_before_running_pipeline(
     runtime_pipeline.run.assert_awaited_once_with(sample_query)
     query_pool.remove_query.assert_awaited_once_with(sample_query)
     session._semaphore.release.assert_called_once_with()
+
+
+@pytest.mark.asyncio
+async def test_controller_schedules_query_without_removing_it_twice(mock_app, sample_query):
+    query_pool = QueryPool()
+    query_pool.queries.append(sample_query)
+    mock_app.query_pool = query_pool
+    mock_app.sess_mgr.get_session = AsyncMock(return_value=SimpleNamespace(_semaphore=asyncio.Semaphore(1)))
+
+    scheduler_errors: list[str] = []
+
+    def stop_on_scheduler_error(message):
+        scheduler_errors.append(str(message))
+        raise asyncio.CancelledError
+
+    def stop_after_scheduling(process_coro, **_kwargs):
+        process_coro.close()
+        raise asyncio.CancelledError
+
+    mock_app.logger.error.side_effect = stop_on_scheduler_error
+    mock_app.task_mgr.create_task.side_effect = stop_after_scheduling
+    controller = Controller(mock_app)
+
+    with pytest.raises(asyncio.CancelledError):
+        await controller.consumer()
+
+    assert scheduler_errors == []
+    assert query_pool.queries == []

tests/unit_tests/pipeline/test_preproc.py

@@ -488,3 +488,115 @@ class TestPreProcessorToolSelection:
         result = await stage.process(query, 'PreProcessor')
 
         assert [tool.name for tool in result.new_query.use_funcs] == ['plugin_tool']
+
+
+class TestPreProcessorDateGrounding:
+    """Tests for current-date injection into the local-agent system prompt."""
+
+    @pytest.mark.asyncio
+    async def test_local_agent_appends_date_to_existing_system_message(self):
+        """Date grounding text should be appended to an existing system prompt."""
+        preproc = get_preproc_module()
+
+        app = FakeApp()
+        mock_session = make_session()
+        app.sess_mgr.get_session = AsyncMock(return_value=mock_session)
+
+        mock_conversation = Mock()
+        mock_conversation.prompt = Mock(messages=[])
+        mock_conversation.prompt.copy = Mock(return_value=Mock(messages=[]))
+        mock_conversation.messages = []
+        mock_conversation.uuid = None
+        app.sess_mgr.get_conversation = AsyncMock(return_value=mock_conversation)
+
+        app.model_mgr.get_model_by_uuid = AsyncMock(return_value=None)
+        app.tool_mgr.get_all_tools = AsyncMock(return_value=[])
+
+        from langbot_plugin.api.entities.builtin.provider import message as provider_message
+
+        system_message = provider_message.Message(role='system', content='You are a helpful assistant.')
+        mock_event_ctx = Mock()
+        mock_event_ctx.event = Mock(default_prompt=[system_message], prompt=[])
+        app.plugin_connector.emit_event = AsyncMock(return_value=mock_event_ctx)
+
+        stage = preproc.PreProcessor(app)
+        query = text_query('hello')
+
+        result = await stage.process(query, 'PreProcessor')
+
+        messages = result.new_query.prompt.messages
+        assert len(messages) == 1
+        assert messages[0].role == 'system'
+        assert messages[0].content.startswith('You are a helpful assistant.')
+        assert 'Current date:' in messages[0].content
+
+    @pytest.mark.asyncio
+    async def test_local_agent_creates_system_message_when_none_exists(self):
+        """A system message should be created when the prompt has none."""
+        preproc = get_preproc_module()
+
+        app = FakeApp()
+        mock_session = make_session()
+        app.sess_mgr.get_session = AsyncMock(return_value=mock_session)
+
+        mock_conversation = Mock()
+        mock_conversation.prompt = Mock(messages=[])
+        mock_conversation.prompt.copy = Mock(return_value=Mock(messages=[]))
+        mock_conversation.messages = []
+        mock_conversation.uuid = None
+        app.sess_mgr.get_conversation = AsyncMock(return_value=mock_conversation)
+
+        app.model_mgr.get_model_by_uuid = AsyncMock(return_value=None)
+        app.tool_mgr.get_all_tools = AsyncMock(return_value=[])
+
+        mock_event_ctx = Mock()
+        mock_event_ctx.event = Mock(default_prompt=[], prompt=[])
+        app.plugin_connector.emit_event = AsyncMock(return_value=mock_event_ctx)
+
+        stage = preproc.PreProcessor(app)
+        query = text_query('hello')
+
+        result = await stage.process(query, 'PreProcessor')
+
+        messages = result.new_query.prompt.messages
+        assert len(messages) == 1
+        assert messages[0].role == 'system'
+        assert 'Current date:' in messages[0].content
+
+    @pytest.mark.asyncio
+    async def test_non_local_agent_runner_skips_date_injection(self):
+        """Runners other than local-agent should not get the date addition."""
+        preproc = get_preproc_module()
+
+        app = FakeApp()
+        mock_session = make_session()
+        app.sess_mgr.get_session = AsyncMock(return_value=mock_session)
+
+        mock_conversation = Mock()
+        mock_conversation.prompt = Mock(messages=[])
+        mock_conversation.prompt.copy = Mock(return_value=Mock(messages=[]))
+        mock_conversation.messages = []
+        mock_conversation.uuid = None
+        app.sess_mgr.get_conversation = AsyncMock(return_value=mock_conversation)
+
+        app.model_mgr.get_model_by_uuid = AsyncMock(return_value=None)
+        app.tool_mgr.get_all_tools = AsyncMock(return_value=[])
+
+        mock_event_ctx = Mock()
+        mock_event_ctx.event = Mock(default_prompt=[], prompt=[])
+        app.plugin_connector.emit_event = AsyncMock(return_value=mock_event_ctx)
+
+        stage = preproc.PreProcessor(app)
+        query = text_query('hello')
+        query.pipeline_config = {
+            'ai': {
+                'runner': {'runner': 'dify-service-api'},
+                'local-agent': {'model': {'primary': '', 'fallbacks': []}, 'prompt': 'default'},
+            },
+            'output': {'misc': {'at-sender': False}},
+            'trigger': {'misc': {}},
+        }
+
+        result = await stage.process(query, 'PreProcessor')
+
+        assert result.new_query.prompt.messages == []

tests/unit_tests/platform/test_websocket_session_isolation.py

@@ -1,12 +1,14 @@
 """Regression tests for isolated embed-widget conversations."""
 
 import asyncio
+import contextvars
 from pathlib import Path
 from unittest.mock import AsyncMock, Mock
 
 import pytest
 
 import langbot_plugin.api.entities.builtin.platform.events as platform_events
+import langbot_plugin.api.entities.builtin.platform.message as platform_message
 from langbot.pkg.platform.sources import websocket_adapter as websocket_adapter_module
 from langbot.pkg.platform.sources.websocket_adapter import WebSocketAdapter, WebSocketMessage, WebSocketSession
 from langbot.pkg.platform.sources.websocket_manager import (
@@ -204,6 +206,48 @@ async def test_embed_event_uses_stable_session_launcher(monkeypatch):
     assert received[0].sender.id == f'websocket_pipeline-1:{session_id}'
 
 
+@pytest.mark.asyncio
+async def test_pipeline_override_survives_detached_listener_task(monkeypatch):
+    manager = WebSocketConnectionManager()
+    connection = await manager.add_connection(
+        websocket=Mock(),
+        scope=SCOPE_A,
+        pipeline_uuid='pipeline-1',
+        session_type='person',
+    )
+    monkeypatch.setattr(websocket_adapter_module, 'ws_connection_manager', manager)
+
+    class DetachedTaskManager:
+        def __init__(self):
+            self.tasks = []
+
+        def create_task(self, coro, **_kwargs):
+            task = asyncio.create_task(coro, context=contextvars.Context())
+            self.tasks.append(task)
+            return Mock(task=task)
+
+    task_manager = DetachedTaskManager()
+    adapter = WebSocketAdapter.model_construct(
+        ap=Mock(task_mgr=task_manager),
+        logger=_adapter_logger(),
+    )
+    adapter.websocket_person_session = WebSocketSession(id='person')
+    adapter.websocket_group_session = WebSocketSession(id='group')
+    pipeline_overrides = []
+
+    async def listener(_event, callback_adapter):
+        pipeline_overrides.append(callback_adapter.get_pipeline_uuid_override())
+
+    adapter.listeners = {platform_events.FriendMessage: listener}
+    await adapter.handle_websocket_message(
+        connection,
+        {'message': [{'type': 'Plain', 'text': 'hello'}], 'stream': False},
+    )
+    await asyncio.gather(*task_manager.tasks)
+
+    assert pipeline_overrides == ['pipeline-1']
+
+
 @pytest.mark.asyncio
 async def test_embed_group_event_uses_stable_session_launcher(monkeypatch):
     manager = WebSocketConnectionManager()
@@ -300,6 +344,49 @@ async def test_stable_session_launcher_resolves_to_active_connection(monkeypatch
     )
 
 
+@pytest.mark.asyncio
+async def test_dashboard_reply_survives_connection_replacement(monkeypatch):
+    manager = WebSocketConnectionManager()
+    original = await manager.add_connection(
+        websocket=Mock(),
+        scope=SCOPE_A,
+        pipeline_uuid='pipeline-1',
+        session_type='person',
+    )
+    monkeypatch.setattr(websocket_adapter_module, 'ws_connection_manager', manager)
+
+    adapter = WebSocketAdapter.model_construct(ap=Mock(), logger=_adapter_logger())
+    adapter.websocket_person_session = WebSocketSession(id='person')
+    adapter.websocket_group_session = WebSocketSession(id='group')
+    received = []
+
+    async def listener(event, _callback_adapter):
+        received.append(event)
+
+    adapter.listeners = {platform_events.FriendMessage: listener}
+    await adapter.handle_websocket_message(
+        original,
+        {'message': [{'type': 'Plain', 'text': 'hello'}], 'stream': False},
+    )
+    await asyncio.sleep(0)
+    await manager.remove_connection(original.connection_id)
+    replacement = await manager.add_connection(
+        websocket=Mock(),
+        scope=SCOPE_A,
+        pipeline_uuid='pipeline-1',
+        session_type='person',
+    )
+
+    await adapter.reply_message(
+        received[0],
+        platform_message.MessageChain([platform_message.Plain(text='done')]),
+    )
+
+    response = await replacement.send_queue.get()
+    assert response['type'] == 'response'
+    assert response['data']['content'] == 'done'
+
+
 def test_session_ids_must_be_canonical_random_uuids():
     assert is_valid_session_id('31c0f2e9-b115-4ee6-8f15-3e624d6456b1')
     assert not is_valid_session_id('session-a')

tests/unit_tests/plugin/test_connector_methods.py

@@ -612,8 +612,13 @@ class TestDisabledPluginEarlyReturns:
         mock_app.instance_config.data = {'plugin': {'enable': False}}
 
         connector = connector_module.PluginRuntimeConnector(mock_app, mock_disconnect)
+        execution_context = connector_module.ExecutionContext(
+            instance_uuid='instance-a',
+            workspace_uuid='workspace-a',
+            placement_generation=1,
+        )
 
-        result = await connector.get_debug_info()
+        result = await connector.get_debug_info(execution_context)
 
         assert result == {}
 

tests/unit_tests/plugin/test_connector_ping.py

@@ -132,6 +132,49 @@ async def test_stdio_runtime_connection_does_not_capture_unconsumed_stderr(
     await connector.aclose()
 
 
+@pytest.mark.asyncio
+async def test_invalid_connect_timeout_is_rejected_before_transport_startup(
+    monkeypatch: pytest.MonkeyPatch,
+):
+    connector = make_connector()
+    connector.ap.instance_config.data['plugin']['connect_timeout_seconds'] = 0
+    stdio_controller = Mock()
+    websocket_controller = Mock()
+    create_task = Mock()
+    get_platform = Mock(return_value='linux')
+    use_websocket = Mock(return_value=False)
+    connector._start_runtime_subprocess = AsyncMock()
+    monkeypatch.setattr(connector_module.constants, 'instance_id', 'instance-a')
+    monkeypatch.setattr(connector_module.asyncio, 'create_task', create_task)
+    monkeypatch.setattr(connector_module.platform, 'get_platform', get_platform)
+    monkeypatch.setattr(
+        connector_module.platform,
+        'use_websocket_to_connect_plugin_runtime',
+        use_websocket,
+    )
+    monkeypatch.setattr(
+        connector_module.stdio_client_controller,
+        'StdioClientController',
+        stdio_controller,
+    )
+    monkeypatch.setattr(
+        connector_module.ws_client_controller,
+        'WebSocketClientController',
+        websocket_controller,
+    )
+
+    with pytest.raises(ValueError, match='plugin.connect_timeout_seconds'):
+        await connector.initialize()
+
+    get_platform.assert_not_called()
+    use_websocket.assert_not_called()
+    stdio_controller.assert_not_called()
+    websocket_controller.assert_not_called()
+    connector._start_runtime_subprocess.assert_not_awaited()
+    create_task.assert_not_called()
+    assert connector._transport_task is None
+
+
 @pytest.mark.asyncio
 async def test_runtime_disconnect_notifies_once_and_clears_handler(
     monkeypatch: pytest.MonkeyPatch,
@@ -282,12 +325,11 @@ def test_closed_deployment_selects_instance_scoped_shared_profile():
     assert connector.runtime_profile == 'shared'
 
 
-def test_external_runtime_control_headers_require_strong_secret(monkeypatch):
+def test_external_runtime_control_headers_are_empty_when_secret_is_unset(monkeypatch):
     monkeypatch.delenv(PLUGIN_RUNTIME_CONTROL_TOKEN_ENV, raising=False)
     connector = make_connector()
 
-    with pytest.raises(PluginRuntimeNotConnectedError, match=PLUGIN_RUNTIME_CONTROL_TOKEN_ENV):
-        connector._control_headers(allow_generate=False)
+    assert connector._control_headers(allow_generate=False) == {}
 
 
 def test_local_runtime_control_headers_generate_ephemeral_secret(monkeypatch):

tests/unit_tests/plugin/test_connector_reconcile.py

@@ -153,6 +153,31 @@ async def test_empty_projected_workspaces_do_not_retain_installation_sets():
     connector.handler.reconcile_plugin_installations.assert_awaited_once_with(())
 
 
+@pytest.mark.asyncio
+async def test_shared_reconcile_logs_workspace_installation_counts_and_elapsed_time():
+    binding_a = execution_binding('workspace-a')
+    binding_b = execution_binding('workspace-b')
+    setting_a = plugin_setting('01', 'a' * 64)
+    setting_b = plugin_setting('02', 'b' * 64)
+    connector = shared_connector(
+        [[binding_a, binding_b]],
+        {'workspace-a': [setting_a], 'workspace-b': [setting_b]},
+    )
+    connector.handler = runtime_handler()
+    await connector._prepare_connected_runtime()
+
+    matching_calls = [
+        call
+        for call in connector.ap.logger.info.call_args_list
+        if call.args
+        and call.args[0]
+        == 'Shared plugin runtime reconcile completed: workspaces=%d desired_installations=%d elapsed_seconds=%.3f'
+    ]
+    assert len(matching_calls) == 1
+    assert matching_calls[0].args[1:3] == (2, 2)
+    assert matching_calls[0].args[3] >= 0
+
+
 @pytest.mark.asyncio
 async def test_fresh_shared_runtime_cache_replays_persisted_local_package():
     package = b'local-lbpkg-bytes'

tests/unit_tests/plugin/test_connector_static.py

@@ -6,9 +6,10 @@ Tests cover:
 
 from __future__ import annotations
 
-import pytest
 from importlib import import_module
 
+import pytest
+
 
 def get_connector_module():
     """Lazy import to avoid circular import issues."""
@@ -60,3 +61,28 @@ def test_runtime_id_is_stable_across_core_restarts(monkeypatch):
     monkeypatch.setattr(connector.constants, 'instance_id', 'instance-a')
 
     assert connector.PluginRuntimeConnector._build_runtime_id() == 'instance-a:plugin-runtime'
+
+
+def test_runtime_connect_timeout_defaults_to_three_minutes():
+    connector = get_connector_module()
+    assert connector.PluginRuntimeConnector._runtime_connect_timeout({}) == 180.0
+
+
+def test_runtime_connect_timeout_reads_typed_plugin_config():
+    connector = get_connector_module()
+    assert connector.PluginRuntimeConnector._runtime_connect_timeout({'connect_timeout_seconds': 45.5}) == 45.5
+
+
+@pytest.mark.parametrize('value', [True, False, None, 0, -1, float('nan'), float('inf'), '180', object()])
+def test_runtime_connect_timeout_rejects_invalid_values(value):
+    connector = get_connector_module()
+    with pytest.raises(ValueError, match='plugin.connect_timeout_seconds'):
+        connector.PluginRuntimeConnector._runtime_connect_timeout({'connect_timeout_seconds': value})
+
+
+def test_runtime_connect_timeout_error_displays_actual_seconds():
+    connector = get_connector_module()
+
+    assert connector.PluginRuntimeConnector._runtime_connect_timeout_error(45.5) == (
+        'Plugin runtime did not become ready within 45.5 seconds'
+    )

tests/unit_tests/provider/test_litellmchat.py

@@ -1304,6 +1304,7 @@ class TestScanModels:
         )
         requester._supports_function_calling = Mock(side_effect=lambda model_id: model_id == 'gpt-4o')
         requester._supports_vision = Mock(side_effect=lambda model_id: model_id == 'gpt-4o')
+        requester._supports_reasoning = Mock(side_effect=lambda model_id: model_id == 'o3')
         requester._safe_context_length = Mock(side_effect=lambda model_id: 128000 if model_id == 'gpt-4o' else None)
 
         mock_response = Mock()
@@ -1311,6 +1312,7 @@ class TestScanModels:
             return_value={
                 'data': [
                     {'id': 'gpt-4o'},
+                    {'id': 'o3'},
                     {'id': 'text-embedding-3-small'},
                     {'id': 'bge-reranker-v2'},
                 ]
@@ -1327,6 +1329,7 @@ class TestScanModels:
         by_id = {model['id']: model for model in result['models']}
         assert by_id['gpt-4o']['abilities'] == ['func_call', 'vision']
         assert by_id['gpt-4o']['context_length'] == 128000
+        assert by_id['o3']['abilities'] == ['reasoning']
         assert by_id['text-embedding-3-small']['type'] == 'embedding'
         assert by_id['bge-reranker-v2']['type'] == 'rerank'
 
@@ -1374,8 +1377,8 @@ class TestScanModels:
         )
 
         with patch.object(litellmchat.litellm, 'get_model_info') as mock_get_model_info:
-            mock_get_model_info.side_effect = (
-                lambda model: {'max_input_tokens': 131072} if model == 'moonshot/moonshot-v1-128k' else {}
+            mock_get_model_info.side_effect = lambda model: (
+                {'max_input_tokens': 131072} if model == 'moonshot/moonshot-v1-128k' else {}
             )
 
             assert requester._safe_context_length('moonshot-v1-128k') == 131072
@@ -1404,8 +1407,8 @@ class TestScanModels:
         )
 
         with patch.object(litellmchat.litellm, 'supports_function_calling') as mock_supports_function_calling:
-            mock_supports_function_calling.side_effect = (
-                lambda model, custom_llm_provider=None: model == 'moonshot/kimi-k2.6' and custom_llm_provider is None
+            mock_supports_function_calling.side_effect = lambda model, custom_llm_provider=None: (
+                model == 'moonshot/kimi-k2.6' and custom_llm_provider is None
             )
 
             assert requester._supports_function_calling('kimi-k2.6') is True

tests/unit_tests/provider/test_localagent_sandbox_exec.py

@@ -178,6 +178,27 @@ def test_stream_accumulator_merges_fragmented_tool_call_arguments():
     assert final_msg.tool_calls[0].function.arguments == '{"command":"pwd"}'
 
 
+def test_stream_accumulator_preserves_tool_call_provider_specific_fields():
+    accumulator = _StreamAccumulator()
+    emitted = accumulator.add(
+        provider_message.MessageChunk(
+            role='assistant',
+            tool_calls=[
+                provider_message.ToolCall(
+                    id='call-gemini',
+                    type='function',
+                    function=provider_message.FunctionCall(name='lookup', arguments='{}'),
+                    provider_specific_fields={'thought_signature': 'sig'},
+                )
+            ],
+            is_final=True,
+        )
+    )
+
+    assert emitted is not None
+    assert emitted.tool_calls[0].provider_specific_fields == {'thought_signature': 'sig'}
+
+
 def test_stream_accumulator_strips_leading_think_from_tool_round_content():
     accumulator = _StreamAccumulator(
         msg_sequence=3,

tests/unit_tests/provider/test_model_service.py

@@ -249,7 +249,11 @@ async def test_updated_llm_model_is_immediately_usable_by_local_agent_pipeline()
         'ai': {
             'runner': {'runner': 'local-agent'},
             'local-agent': {
-                'model': {'primary': model_uuid, 'fallbacks': []},
+                'model': {
+                    'primary': model_uuid,
+                    'fallbacks': [],
+                    'reasoning': {model_uuid: 'high'},
+                },
                 'prompt': [],
                 'knowledge-bases': [],
             },
@@ -293,3 +297,134 @@ async def test_updated_llm_model_is_immediately_usable_by_local_agent_pipeline()
     candidates = await LocalAgentRunner._get_model_candidates(runner, processed_query)
 
     assert [model.model_entity.uuid for model in candidates] == [model_uuid]
+    assert candidates[0].reasoning_config_override == {'level': 'high'}
+
+
+@pytest.mark.asyncio
+async def test_local_agent_applies_reasoning_per_fallback_model():
+    execution_context = ExecutionContext(
+        instance_uuid='instance-test',
+        workspace_uuid='workspace-test',
+        placement_generation=1,
+    )
+    provider = Mock(
+        execution_context=execution_context,
+        provider_entity=persistence_model.ModelProvider(
+            workspace_uuid='workspace-test',
+            uuid='provider',
+            name='provider',
+            requester='openai',
+            base_url='https://example.com',
+            api_keys=[],
+        ),
+    )
+    primary = requester.RuntimeLLMModel(
+        execution_context,
+        persistence_model.LLMModel(
+            workspace_uuid='workspace-test',
+            uuid='primary-model',
+            name='primary',
+            provider_uuid='provider',
+            abilities=['reasoning'],
+            extra_args={},
+        ),
+        provider,
+    )
+    fallback = requester.RuntimeLLMModel(
+        execution_context,
+        persistence_model.LLMModel(
+            workspace_uuid='workspace-test',
+            uuid='fallback-model',
+            name='fallback',
+            provider_uuid='provider',
+            abilities=['reasoning'],
+            extra_args={},
+        ),
+        provider,
+    )
+    models = {'primary-model': primary, 'fallback-model': fallback}
+    runner = SimpleNamespace(
+        ap=SimpleNamespace(
+            model_mgr=SimpleNamespace(
+                get_model_by_uuid=AsyncMock(side_effect=lambda _context, model_uuid: models[model_uuid]),
+            ),
+            logger=Mock(),
+        )
+    )
+    query = SimpleNamespace(
+        use_llm_model_uuid='primary-model',
+        variables={'_fallback_model_uuids': ['fallback-model']},
+        pipeline_config={
+            'ai': {
+                'local-agent': {
+                    'model': {
+                        'primary': 'primary-model',
+                        'fallbacks': ['fallback-model'],
+                        'reasoning': {
+                            'primary-model': 'low',
+                            'fallback-model': 'high',
+                        },
+                    }
+                }
+            }
+        },
+        _execution_context=execution_context,
+    )
+
+    candidates = await LocalAgentRunner._get_model_candidates(runner, query)
+
+    assert [candidate.reasoning_config_override for candidate in candidates] == [
+        {'level': 'low'},
+        {'level': 'high'},
+    ]
+    assert candidates[0] is not primary
+    assert candidates[1] is not fallback
+    assert primary.reasoning_config_override is None
+    assert fallback.reasoning_config_override is None
+
+
+def test_local_agent_rejects_invalid_pipeline_reasoning_level():
+    execution_context = ExecutionContext(
+        instance_uuid='instance-test',
+        workspace_uuid='workspace-test',
+        placement_generation=1,
+    )
+    provider = Mock(
+        execution_context=execution_context,
+        provider_entity=persistence_model.ModelProvider(
+            workspace_uuid='workspace-test',
+            uuid='provider',
+            name='provider',
+            requester='openai',
+            base_url='https://example.com',
+            api_keys=[],
+        ),
+    )
+    model = requester.RuntimeLLMModel(
+        execution_context,
+        persistence_model.LLMModel(
+            workspace_uuid='workspace-test',
+            uuid='primary-model',
+            name='primary',
+            provider_uuid='provider',
+            abilities=['reasoning'],
+            extra_args={},
+        ),
+        provider,
+    )
+    query = SimpleNamespace(
+        pipeline_config={
+            'ai': {
+                'local-agent': {
+                    'model': {
+                        'primary': 'primary-model',
+                        'fallbacks': [],
+                        'reasoning': {'primary-model': 'turbo'},
+                    }
+                }
+            }
+        }
+    )
+
+    with pytest.raises(ValueError, match='Unsupported reasoning level'):
+        LocalAgentRunner._apply_pipeline_reasoning_config(query, model)

tests/unit_tests/provider/test_reasoning_control.py

@@ -0,0 +1,872 @@
+from __future__ import annotations
+
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import pytest
+
+import langbot_plugin.api.entities.builtin.provider.message as provider_message
+from langbot.pkg.api.http.context import ExecutionContext
+from langbot.pkg.entity.persistence import model as persistence_model
+from langbot.pkg.provider.modelmgr import errors, reasoning, requester
+from langbot.pkg.provider.modelmgr.requesters import litellmchat
+from langbot.pkg.provider.modelmgr.requesters.litellmchat import LiteLLMRequester
+from langbot.pkg.provider.runners.localagent import _StreamAccumulator
+
+
+def _runtime_model(
+    request: LiteLLMRequester,
+    level: str = 'provider_default',
+    name: str = 'reasoning-model',
+    abilities: list[str] | None = None,
+    requester_name: str | None = None,
+) -> requester.RuntimeLLMModel:
+    execution_context = ExecutionContext(
+        instance_uuid='instance-test',
+        workspace_uuid='workspace-test',
+        placement_generation=1,
+    )
+    entity = persistence_model.LLMModel(
+        workspace_uuid='workspace-test',
+        uuid='reasoning-model',
+        name=name,
+        provider_uuid='provider-test',
+        abilities=abilities if abilities is not None else ['reasoning'],
+        reasoning_config={'level': level},
+        extra_args={},
+    )
+    provider = SimpleNamespace(
+        execution_context=execution_context,
+        provider_entity=persistence_model.ModelProvider(
+            workspace_uuid='workspace-test',
+            uuid='provider-test',
+            name='provider',
+            requester=requester_name or request.requester_cfg.get('requester_name') or 'custom-requester',
+            base_url='https://example.com',
+            api_keys=[],
+        ),
+        requester=request,
+        token_mgr=SimpleNamespace(),
+    )
+    return requester.RuntimeLLMModel(execution_context, entity, provider)
+
+
+def _requester(provider: str = '', requester_name: str = '') -> LiteLLMRequester:
+    return LiteLLMRequester(
+        SimpleNamespace(),
+        {
+            'custom_llm_provider': provider,
+            'requester_name': requester_name,
+        },
+    )
+
+
+def test_reasoning_config_normalization_and_conflicts():
+    assert reasoning.normalize_reasoning_config(None) == {'level': 'provider_default'}
+    assert reasoning.normalize_reasoning_config({}) == {'level': 'provider_default'}
+    assert reasoning.validate_reasoning_config(
+        {'level': 'high'},
+        ['reasoning'],
+        {},
+    ) == {'level': 'high'}
+
+    with pytest.raises(ValueError, match='Unsupported reasoning level'):
+        reasoning.normalize_reasoning_config({'level': 'turbo'})
+    with pytest.raises(ValueError, match='reasoning ability'):
+        reasoning.validate_reasoning_config({'level': 'low'}, [], {})
+    with pytest.raises(ValueError, match='extra_body.thinking_budget'):
+        reasoning.validate_reasoning_config(
+            {'level': 'low'},
+            ['reasoning'],
+            {'extra_body': {'thinking_budget': 1024}},
+        )
+    assert reasoning.find_reasoning_arg_conflicts(
+        {
+            'enable_thinking': True,
+            'extra_body': {'reasoning_effort': 'high'},
+        }
+    ) == ['enable_thinking', 'extra_body.reasoning_effort']
+
+
+def test_manual_reasoning_model_without_known_protocol_stays_conservative(monkeypatch):
+    request = _requester()
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request))
+
+    assert capabilities == {
+        'supported': True,
+        'levels': ['provider_default'],
+        'source': 'manual',
+    }
+
+
+def test_openai_protocol_does_not_mark_unknown_models_as_reasoning(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(
+        _runtime_model(request, name='future-reasoning-model', abilities=[])
+    )
+
+    assert capabilities == {
+        'supported': False,
+        'levels': ['provider_default'],
+        'source': 'unknown',
+    }
+
+
+def test_unknown_unmarked_model_without_provider_stays_safe(monkeypatch):
+    request = _requester()
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name='unknown-model', abilities=[]))
+
+    assert capabilities == {
+        'supported': False,
+        'levels': ['provider_default'],
+        'source': 'unknown',
+    }
+
+
+def test_mimo_exposes_off_on_without_fake_effort_levels(monkeypatch):
+    request = _requester('openai', 'mimo-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name='mimo-v2.5', abilities=[]))
+
+    assert capabilities == {
+        'supported': True,
+        'levels': ['provider_default', 'disabled', 'enabled'],
+        'source': 'provider',
+    }
+
+
+def test_openai_reasoning_levels_follow_litellm_metadata(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(
+        request,
+        '_safe_model_info',
+        lambda _: {
+            'supports_none_reasoning_effort': True,
+            'supports_minimal_reasoning_effort': False,
+            'supports_low_reasoning_effort': True,
+            'supports_xhigh_reasoning_effort': True,
+        },
+    )
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name='gpt-5'))
+
+    assert capabilities['source'] == 'litellm'
+    assert capabilities['levels'] == [
+        'provider_default',
+        'disabled',
+        'low',
+        'medium',
+        'high',
+        'xhigh',
+    ]
+
+
+def test_anthropic_adaptive_and_always_on_profiles(monkeypatch):
+    request = _requester('anthropic', 'anthropic-messages')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    adaptive = request.get_reasoning_capabilities(_runtime_model(request, name='claude-sonnet-4-6', abilities=[]))
+    assert adaptive['levels'] == [
+        'provider_default',
+        'disabled',
+        'low',
+        'medium',
+        'high',
+        'xhigh',
+        'max',
+    ]
+
+    always_on = request.get_reasoning_capabilities(_runtime_model(request, name='claude-fable-5', abilities=[]))
+    assert 'disabled' not in always_on['levels']
+
+    legacy = request.get_reasoning_capabilities(_runtime_model(request, name='claude-3-5-sonnet', abilities=[]))
+    assert legacy['levels'] == ['provider_default', 'low', 'medium', 'high']
+
+
+def test_deepseek_profiles_match_model_generation(monkeypatch):
+    request = _requester('deepseek', 'deepseek-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    assert request.get_reasoning_capabilities(_runtime_model(request, name='deepseek-v4-flash', abilities=[]))[
+        'levels'
+    ] == ['provider_default', 'disabled', 'low', 'high', 'xhigh', 'max']
+    assert request.get_reasoning_capabilities(_runtime_model(request, name='deepseek-chat', abilities=[]))[
+        'levels'
+    ] == ['provider_default', 'disabled', 'enabled']
+    assert request.get_reasoning_capabilities(_runtime_model(request, name='deepseek-r1', abilities=[]))['levels'] == [
+        'provider_default'
+    ]
+
+
+@pytest.mark.parametrize(
+    ('model_name', 'expected_levels'),
+    [
+        ('kimi-k3', ['provider_default', 'low', 'high', 'max']),
+        ('kimi-k2.7-code', ['provider_default']),
+        ('kimi-k2.6', ['provider_default', 'disabled', 'enabled']),
+        ('kimi-k2.5', ['provider_default', 'disabled', 'enabled']),
+    ],
+)
+def test_kimi_profiles(model_name, expected_levels, monkeypatch):
+    request = _requester('openai', 'moonshot-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name=model_name, abilities=[]))
+
+    assert capabilities['levels'] == expected_levels
+
+
+def test_qwen_mixed_and_dedicated_thinking_profiles(monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    mixed = request.get_reasoning_capabilities(_runtime_model(request, name='qwen-plus', abilities=[]))
+    dedicated = request.get_reasoning_capabilities(
+        _runtime_model(request, name='qwen3-235b-a22b-thinking-2507', abilities=[])
+    )
+
+    assert mixed['levels'] == ['provider_default', 'disabled', 'enabled']
+    assert dedicated['levels'] == ['provider_default', 'low', 'medium', 'high']
+
+
+def test_qwen3_exposes_budget_based_reasoning_levels(monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    mixed = request.get_reasoning_capabilities(_runtime_model(request, name='qwen3.8-max', abilities=[]))
+    dedicated = request.get_reasoning_capabilities(
+        _runtime_model(request, name='qwen3.7-max-preview', abilities=[])
+    )
+
+    assert mixed['levels'] == ['provider_default', 'disabled', 'low', 'medium', 'high']
+    assert mixed['legacy_levels'] == ['enabled']
+    assert dedicated['levels'] == ['provider_default', 'low', 'medium', 'high']
+
+
+def test_qwen3_legacy_enabled_config_remains_supported(monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    assert request._build_reasoning_args(_runtime_model(request, 'enabled', name='qwen3.8-max')) == {
+        'extra_body': {'enable_thinking': True}
+    }
+
+
+@pytest.mark.parametrize(
+    ('level', 'budget'),
+    [('low', 1024), ('medium', 4096), ('high', 8192)],
+)
+def test_qwen3_reasoning_levels_translate_to_thinking_budget(level, budget, monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    assert request._build_reasoning_args(_runtime_model(request, level, name='qwen3.8-max')) == {
+        'extra_body': {
+            'enable_thinking': True,
+            'thinking_budget': budget,
+        }
+    }
+
+
+@pytest.mark.parametrize('model_name', ['qwen3.7-max-preview', 'qwen3.7-max-2026-05-17'])
+def test_qwen_dedicated_thinking_release_models_are_not_toggleable(model_name, monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name=model_name, abilities=[]))
+
+    assert capabilities['levels'] == ['provider_default', 'low', 'medium', 'high']
+
+
+@pytest.mark.parametrize(
+    ('model_name', 'expected_levels'),
+    [
+        ('kimi-k2.6', ['provider_default', 'disabled', 'enabled']),
+        ('kimi-k2.5', ['provider_default', 'disabled', 'enabled']),
+        ('kimi-k2.7-code', ['provider_default']),
+        ('kimi-k2-thinking', ['provider_default']),
+    ],
+)
+def test_bailian_kimi_profiles_use_kimi_model_rules(model_name, expected_levels, monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name=model_name, abilities=[]))
+
+    assert capabilities['levels'] == expected_levels
+
+
+def test_bailian_kimi_uses_thinking_protocol_instead_of_qwen_protocol(monkeypatch):
+    request = _requester('openai', 'bailian-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    assert request._build_reasoning_args(_runtime_model(request, 'disabled', name='kimi-k2.6')) == {
+        'extra_body': {'thinking': {'type': 'disabled'}}
+    }
+
+
+def test_doubao_exposes_documented_effort_range(monkeypatch):
+    request = _requester('openai', 'doubao-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(
+        _runtime_model(request, name='doubao-seed-2-1-pro-260628', abilities=[])
+    )
+
+    assert capabilities['levels'] == ['provider_default', 'disabled', 'low', 'medium', 'high']
+
+
+@pytest.mark.parametrize(
+    ('model_name', 'expected_levels'),
+    [
+        ('gpt-5', ['provider_default', 'low', 'medium', 'high']),
+        (
+            'claude-sonnet-4-6',
+            ['provider_default', 'disabled', 'low', 'medium', 'high', 'xhigh', 'max'],
+        ),
+        ('deepseek-v4-flash', ['provider_default', 'disabled', 'low', 'high', 'xhigh', 'max']),
+        ('kimi-k2.6', ['provider_default', 'disabled', 'enabled']),
+        ('qwen-plus', ['provider_default', 'disabled', 'enabled']),
+        ('doubao-seed-2-1-pro-260628', ['provider_default', 'disabled', 'low', 'medium', 'high']),
+        ('mimo-v2.5', ['provider_default', 'disabled', 'enabled']),
+    ],
+)
+def test_new_api_infers_upstream_protocol_from_model_name(model_name, expected_levels, monkeypatch):
+    request = _requester('openai', 'new-api-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(request, '_safe_model_info', lambda _: {})
+
+    capabilities = request.get_reasoning_capabilities(_runtime_model(request, name=model_name, abilities=[]))
+
+    assert capabilities['levels'] == expected_levels
+
+
+@pytest.mark.parametrize(
+    ('provider', 'requester_name', 'model_name'),
+    [
+        ('openai', 'openai-chat-completions', 'gpt-5'),
+        ('anthropic', 'anthropic-messages', 'claude-sonnet-4-6'),
+        ('deepseek', 'deepseek-chat-completions', 'deepseek-v4-flash'),
+        ('openai', 'mimo-chat-completions', 'mimo-v2.5'),
+        ('openai', 'moonshot-chat-completions', 'kimi-k2.6'),
+        ('openai', 'bailian-chat-completions', 'qwen-plus'),
+        ('openai', 'doubao-chat-completions', 'doubao-seed-2-1-pro-260628'),
+        ('openai', 'new-api-chat-completions', 'deepseek-v4-flash'),
+    ],
+)
+def test_scanned_known_models_gain_reasoning_ability(provider, requester_name, model_name, monkeypatch):
+    request = _requester(provider, requester_name)
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(request, '_supports_function_calling', lambda _: False)
+    monkeypatch.setattr(request, '_supports_vision', lambda _: False)
+    monkeypatch.setattr(request, '_safe_context_length', lambda _: None)
+
+    scanned = request._enrich_scanned_model(model_name)
+
+    assert scanned['abilities'] == ['reasoning']
+
+
+def test_new_api_unknown_alias_stays_conservative(monkeypatch):
+    request = _requester('openai', 'new-api-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+
+    capabilities = request.get_reasoning_capabilities(
+        _runtime_model(request, name='company-internal-alias', abilities=[])
+    )
+
+    assert capabilities == {
+        'supported': False,
+        'levels': ['provider_default'],
+        'source': 'unknown',
+    }
+
+
+def test_reasoning_argument_translation(monkeypatch):
+    openai_request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(openai_request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(openai_request, '_safe_model_info', lambda _: {'supports_none_reasoning_effort': True})
+    assert openai_request._build_reasoning_args(_runtime_model(openai_request, 'disabled', name='gpt-5')) == {
+        'reasoning_effort': 'none'
+    }
+
+    anthropic_request = _requester('anthropic', 'anthropic-messages')
+    assert anthropic_request._build_reasoning_args(
+        _runtime_model(anthropic_request, 'disabled', name='claude-sonnet-4-6')
+    ) == {'thinking': {'type': 'disabled'}}
+
+    deepseek_request = _requester('deepseek', 'deepseek-chat-completions')
+    assert deepseek_request._build_reasoning_args(
+        _runtime_model(deepseek_request, 'high', name='deepseek-v4-flash')
+    ) == {
+        'extra_body': {
+            'thinking': {'type': 'enabled'},
+            'reasoning_effort': 'high',
+        }
+    }
+
+    kimi_request = _requester('openai', 'moonshot-chat-completions')
+    assert kimi_request._build_reasoning_args(_runtime_model(kimi_request, 'enabled', name='kimi-k2.6')) == {
+        'extra_body': {'thinking': {'type': 'enabled'}}
+    }
+    assert kimi_request._build_reasoning_args(_runtime_model(kimi_request, 'high', name='kimi-k3')) == {
+        'reasoning_effort': 'high'
+    }
+
+    qwen_request = _requester('openai', 'bailian-chat-completions')
+    assert qwen_request._build_reasoning_args(_runtime_model(qwen_request, 'disabled', name='qwen-plus')) == {
+        'extra_body': {'enable_thinking': False}
+    }
+
+    doubao_request = _requester('openai', 'doubao-chat-completions')
+    assert doubao_request._build_reasoning_args(
+        _runtime_model(doubao_request, 'high', name='doubao-seed-2-1-pro-260628')
+    ) == {'reasoning_effort': 'high'}
+
+    mimo_request = _requester('openai', 'mimo-chat-completions')
+    assert mimo_request._build_reasoning_args(_runtime_model(mimo_request, 'disabled', name='mimo-v2.5')) == {
+        'extra_body': {'thinking': {'type': 'disabled'}}
+    }
+
+
+def test_pipeline_reasoning_override_takes_precedence(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(request, '_safe_model_info', lambda _: {})
+    model = _runtime_model(request, 'high', name='gpt-5')
+
+    model.reasoning_config_override = {'level': 'provider_default'}
+    assert request._build_reasoning_args(model) == {}
+
+    model.reasoning_config_override = {'level': 'low'}
+    assert request._build_reasoning_args(model) == {'reasoning_effort': 'low'}
+
+
+def test_always_on_reasoning_models_do_not_offer_disabled(monkeypatch):
+    deepseek_request = _requester('deepseek', 'deepseek-chat-completions')
+    monkeypatch.setattr(deepseek_request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(deepseek_request, '_safe_model_info', lambda _: {})
+    deepseek_capabilities = deepseek_request.get_reasoning_capabilities(
+        _runtime_model(deepseek_request, name='deepseek-r1')
+    )
+    assert deepseek_capabilities['levels'] == ['provider_default']
+
+    gemini_request = _requester('gemini')
+    monkeypatch.setattr(gemini_request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(
+        gemini_request,
+        '_safe_model_info',
+        lambda _: {'supports_none_reasoning_effort': True},
+    )
+    gemini_capabilities = gemini_request.get_reasoning_capabilities(_runtime_model(gemini_request, name='gemini-3-pro'))
+    assert 'disabled' not in gemini_capabilities['levels']
+    with pytest.raises(errors.RequesterError, match='not supported'):
+        gemini_request._build_reasoning_args(_runtime_model(gemini_request, 'disabled', name='gemini-3-pro'))
+
+
+def test_non_target_provider_capabilities_remain_supported(monkeypatch):
+    ollama_request = _requester('ollama', 'ollama')
+    monkeypatch.setattr(ollama_request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(ollama_request, '_safe_model_info', lambda _: {})
+
+    toggle_capabilities = ollama_request.get_reasoning_capabilities(_runtime_model(ollama_request, name='qwen3'))
+    assert toggle_capabilities['levels'] == [
+        'provider_default',
+        'disabled',
+        'enabled',
+    ]
+    assert ollama_request._build_reasoning_args(_runtime_model(ollama_request, 'enabled', name='qwen3')) == {
+        'reasoning_effort': 'low'
+    }
+
+    effort_capabilities = ollama_request.get_reasoning_capabilities(_runtime_model(ollama_request, name='gpt-oss:20b'))
+    assert effort_capabilities['levels'] == [
+        'provider_default',
+        'disabled',
+        'low',
+        'medium',
+        'high',
+    ]
+    assert ollama_request._build_reasoning_args(_runtime_model(ollama_request, 'high', name='gpt-oss:20b')) == {
+        'reasoning_effort': 'high'
+    }
+
+    volcengine_request = _requester('volcengine', 'volcark-chat-completions')
+    monkeypatch.setattr(volcengine_request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(volcengine_request, '_safe_model_info', lambda _: {})
+    assert volcengine_request._build_reasoning_args(
+        _runtime_model(volcengine_request, 'disabled', name='doubao-seed')
+    ) == {'extra_body': {'thinking': {'type': 'disabled'}}}
+
+
+def test_explicit_unsupported_level_raises(monkeypatch):
+    request = _requester()
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(request, '_safe_model_info', lambda _: {})
+
+    with pytest.raises(errors.RequesterError, match='Available levels: provider_default'):
+        request._build_reasoning_args(_runtime_model(request, 'high', abilities=[]))
+
+
+def test_provider_inference_rejects_levels_outside_conservative_profile(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+    monkeypatch.setattr(request, '_safe_model_info', lambda _: {})
+
+    with pytest.raises(errors.RequesterError, match='Available levels: provider_default, low, medium, high'):
+        request._build_reasoning_args(_runtime_model(request, 'xhigh', name='gpt-5', abilities=[]))
+
+
+@pytest.mark.asyncio
+async def test_completion_args_reject_reasoning_extra_arg_conflicts(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: True)
+    monkeypatch.setattr(request, '_safe_model_info', lambda _: {})
+    model = _runtime_model(request, 'high', name='gpt-5')
+    model.model_entity.extra_args = {'reasoning_effort': 'low'}
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+
+    with pytest.raises(errors.RequesterError, match='conflicts with advanced parameters'):
+        await request._build_completion_args(model, [])
+
+
+@pytest.mark.asyncio
+async def test_openai_compatible_reasoning_effort_is_explicitly_allowed(monkeypatch):
+    request = _requester('openai', 'moonshot-chat-completions')
+    model = _runtime_model(request, 'high', name='kimi-k3')
+    model.model_entity.extra_args = {'allowed_openai_params': ['custom_extension']}
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+
+    args = await request._build_completion_args(model, [])
+
+    assert args['reasoning_effort'] == 'high'
+    assert args['allowed_openai_params'] == ['custom_extension', 'reasoning_effort']
+
+
+@pytest.mark.asyncio
+async def test_provider_default_does_not_allow_or_send_reasoning_effort():
+    request = _requester('openai', 'new-api-chat-completions')
+    model = _runtime_model(request, 'provider_default', name='deepseek-v4-flash')
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+
+    args = await request._build_completion_args(model, [])
+
+    assert 'reasoning_effort' not in args
+    assert 'allowed_openai_params' not in args
+
+
+@pytest.mark.asyncio
+async def test_deepseek_disabled_thinking_is_merged_into_extra_body(monkeypatch):
+    request = _requester('deepseek', 'deepseek-chat-completions')
+    monkeypatch.setattr(request, '_supports_reasoning', lambda _: False)
+    model = _runtime_model(request, 'disabled', name='deepseek-chat')
+    model.model_entity.extra_args = {'extra_body': {'custom_extension': True}}
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+
+    args = await request._build_completion_args(model, [])
+
+    assert args['extra_body'] == {
+        'custom_extension': True,
+        'thinking': {'type': 'disabled'},
+    }
+
+
+@pytest.mark.asyncio
+async def test_openai_compatible_reasoning_history_is_promoted_for_tool_continuity():
+    request = _requester('openai', 'mimo-chat-completions')
+    model = _runtime_model(request, 'enabled', name='mimo-v2.5')
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+    history = [
+        provider_message.Message(
+            role='assistant',
+            content='<think>\nprior reasoning\n</think>\nanswer',
+            provider_specific_fields={'reasoning_content': 'prior reasoning'},
+        )
+    ]
+
+    args = await request._build_completion_args(model, history)
+
+    assert args['messages'][0]['reasoning_content'] == 'prior reasoning'
+    assert args['messages'][0]['content'] == 'answer'
+    assert 'provider_specific_fields' not in args['messages'][0]
+
+
+@pytest.mark.asyncio
+async def test_disabling_reasoning_removes_previous_reasoning_context():
+    request = _requester('openai', 'mimo-chat-completions')
+    model = _runtime_model(request, 'disabled', name='mimo-v2.5')
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+    history = [
+        provider_message.Message(
+            role='assistant',
+            content='answer',
+            provider_specific_fields={'reasoning_content': 'prior reasoning'},
+        )
+    ]
+
+    args = await request._build_completion_args(model, history)
+
+    assert 'reasoning_content' not in args['messages'][0]
+    assert 'provider_specific_fields' not in args['messages'][0]
+
+
+@pytest.mark.asyncio
+async def test_anthropic_history_promotes_thinking_blocks_instead_of_reasoning_content():
+    request = _requester('anthropic', 'anthropic-messages')
+    model = _runtime_model(request, 'high', name='claude-sonnet-4-6')
+    model.provider.token_mgr.get_token = lambda: 'test-token'
+    thinking_blocks = [{'type': 'thinking', 'thinking': 'prior reasoning', 'signature': 'sig'}]
+    history = [
+        provider_message.Message(
+            role='assistant',
+            content='',
+            provider_specific_fields={
+                'reasoning_content': 'prior reasoning',
+                'thinking_blocks': thinking_blocks,
+            },
+        )
+    ]
+
+    args = await request._build_completion_args(model, history)
+
+    assert args['messages'][0]['thinking_blocks'] == thinking_blocks
+    assert 'reasoning_content' not in args['messages'][0]
+    assert 'provider_specific_fields' not in args['messages'][0]
+
+
+@pytest.mark.asyncio
+async def test_non_stream_anthropic_thinking_blocks_are_preserved(monkeypatch):
+    request = _requester('anthropic', 'anthropic-messages')
+    request._build_completion_args = AsyncMock(return_value={})
+    thinking_blocks = [{'type': 'thinking', 'thinking': 'private reasoning', 'signature': 'sig'}]
+    response = SimpleNamespace(
+        choices=[
+            SimpleNamespace(
+                message=_Dumpable(
+                    {
+                        'role': 'assistant',
+                        'content': 'answer',
+                        'thinking_blocks': thinking_blocks,
+                    }
+                )
+            )
+        ],
+        usage=None,
+    )
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=response))
+
+    message, _ = await request.invoke_llm(None, _runtime_model(request, 'high', name='claude-sonnet-4-6'), [])
+
+    assert message.content == '<think>\nprivate reasoning\n</think>\nanswer'
+    assert message.provider_specific_fields == {'thinking_blocks': thinking_blocks}
+
+
+class _Dumpable:
+    def __init__(self, data: dict):
+        self.data = data
+
+    def model_dump(self) -> dict:
+        return dict(self.data)
+
+
+@pytest.mark.asyncio
+async def test_non_stream_reasoning_content_is_preserved(monkeypatch):
+    request = _requester('deepseek')
+    request._build_completion_args = AsyncMock(return_value={})
+    response = SimpleNamespace(
+        choices=[
+            SimpleNamespace(
+                message=_Dumpable(
+                    {
+                        'role': 'assistant',
+                        'content': 'answer',
+                        'reasoning_content': 'private reasoning',
+                    }
+                )
+            )
+        ],
+        usage=None,
+    )
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=response))
+
+    message, _ = await request.invoke_llm(None, _runtime_model(request), [], remove_think=True)
+
+    assert message.content == 'answer'
+    assert message.provider_specific_fields == {'reasoning_content': 'private reasoning'}
+
+
+@pytest.mark.asyncio
+async def test_stream_reasoning_round_trip_with_hidden_display(monkeypatch):
+    request = _requester('deepseek')
+    request._build_completion_args = AsyncMock(return_value={})
+
+    async def chunks():
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'role': 'assistant', 'reasoning_content': 'private '}),
+                    finish_reason=None,
+                )
+            ],
+            usage=None,
+        )
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'content': 'answer'}),
+                    finish_reason='stop',
+                )
+            ],
+            usage=None,
+        )
+
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=chunks()))
+    accumulator = _StreamAccumulator(remove_think=True)
+    emitted: provider_message.MessageChunk | None = None
+
+    async for chunk in request.invoke_llm_stream(
+        None,
+        _runtime_model(request),
+        [],
+        remove_think=True,
+    ):
+        emitted = accumulator.add(chunk) or emitted
+
+    assert emitted is not None
+    assert emitted.content == 'answer'
+    assert emitted.provider_specific_fields == {'reasoning_content': 'private '}
+
+
+@pytest.mark.asyncio
+async def test_stream_reasoning_content_is_wrapped_for_display(monkeypatch):
+    request = _requester('deepseek')
+    request._build_completion_args = AsyncMock(return_value={})
+
+    async def chunks():
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'role': 'assistant', 'reasoning_content': 'private '}),
+                    finish_reason=None,
+                )
+            ],
+            usage=None,
+        )
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'content': 'answer'}),
+                    finish_reason='stop',
+                )
+            ],
+            usage=None,
+        )
+
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=chunks()))
+    accumulator = _StreamAccumulator(remove_think=False)
+    emitted: provider_message.MessageChunk | None = None
+
+    async for chunk in request.invoke_llm_stream(
+        None,
+        _runtime_model(request),
+        [],
+        remove_think=False,
+    ):
+        emitted = accumulator.add(chunk) or emitted
+
+    assert emitted is not None
+    assert emitted.content == '<think>\nprivate \n</think>\nanswer'
+    assert emitted.provider_specific_fields == {'reasoning_content': 'private '}
+
+
+@pytest.mark.asyncio
+async def test_stream_anthropic_thinking_blocks_are_preserved(monkeypatch):
+    request = _requester('anthropic', 'anthropic-messages')
+    request._build_completion_args = AsyncMock(return_value={})
+    thinking_blocks = [{'type': 'thinking', 'thinking': 'private ', 'signature': 'sig'}]
+
+    async def chunks():
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'role': 'assistant', 'thinking_blocks': thinking_blocks}),
+                    finish_reason=None,
+                )
+            ],
+            usage=None,
+        )
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable({'content': 'answer'}),
+                    finish_reason='stop',
+                )
+            ],
+            usage=None,
+        )
+
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=chunks()))
+    accumulator = _StreamAccumulator(remove_think=False)
+    emitted: provider_message.MessageChunk | None = None
+
+    async for chunk in request.invoke_llm_stream(
+        None,
+        _runtime_model(request, 'high', name='claude-sonnet-4-6'),
+        [],
+        remove_think=False,
+    ):
+        emitted = accumulator.add(chunk) or emitted
+
+    assert emitted is not None
+    assert emitted.content == '<think>\nprivate \n</think>\nanswer'
+    assert emitted.provider_specific_fields == {'thinking_blocks': thinking_blocks}
+
+
+@pytest.mark.asyncio
+async def test_hidden_thinking_does_not_drop_same_delta_tool_call(monkeypatch):
+    request = _requester('openai', 'openai-chat-completions')
+    request._build_completion_args = AsyncMock(return_value={})
+
+    async def chunks():
+        yield SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    delta=_Dumpable(
+                        {
+                            'content': '<think>hidden</think>',
+                            'tool_calls': [
+                                {
+                                    'index': 0,
+                                    'id': 'call_1',
+                                    'type': 'function',
+                                    'function': {'name': 'lookup', 'arguments': '{}'},
+                                }
+                            ],
+                        }
+                    ),
+                    finish_reason='tool_calls',
+                )
+            ],
+            usage=None,
+        )
+
+    monkeypatch.setattr(litellmchat, 'acompletion', AsyncMock(return_value=chunks()))
+    collected = [
+        chunk
+        async for chunk in request.invoke_llm_stream(
+            None,
+            _runtime_model(request, 'provider_default'),
+            [],
+            remove_think=True,
+        )
+    ]
+
+    assert len(collected) == 1
+    assert collected[0].tool_calls[0].id == 'call_1'

tests/unit_tests/provider/test_requester_base.py

@@ -400,6 +400,7 @@ def test_runtime_llm_model_initialization(runtime_llm_model, fake_persistence_da
     assert model.model_entity.abilities == model_entity.abilities
     assert model.model_entity.extra_args == model_entity.extra_args
     assert model.provider is not None
+    assert model.reasoning_config_override is None
 
 
 def test_runtime_llm_model_provider_ref(runtime_llm_model):

tests/unit_tests/telemetry/test_heartbeat.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import json
+from datetime import datetime, timezone
 from types import SimpleNamespace
 
 import pytest
@@ -14,6 +15,12 @@ def get_heartbeat_module():
     return import_module('langbot.pkg.telemetry.heartbeat')
 
 
+def test_workspace_created_timestamp_treats_naive_database_values_as_utc():
+    heartbeat = get_heartbeat_module()
+    created_at = datetime(2026, 8, 4, 0, 0, 0)
+    assert heartbeat._workspace_created_timestamp(created_at) == 1785801600
+
+
 def make_app():
     ap = Mock()
     ap.instance_config = Mock()
@@ -57,15 +64,17 @@ def make_app():
 
 class TestBuildHeartbeatPayload:
     @pytest.mark.asyncio
-    async def test_payload_shape(self):
+    async def test_payload_shape(self, monkeypatch):
         heartbeat = get_heartbeat_module()
+        monkeypatch.setattr(heartbeat.constants, 'instance_id', 'instance-test')
         ap = make_app()
         payload = await heartbeat.build_heartbeat_payload(ap, workspace_uuid='workspace-a')
 
         assert payload['event_type'] == 'instance_heartbeat'
         assert payload['query_id'] == ''
         assert payload['workspace_uuid'] == 'workspace-a'
-        assert 'instance_id' not in payload
+        assert payload['instance_id']
+        assert payload['workspace_create_ts'] == 0
         assert 'instance_create_ts' in payload
         assert 'timestamp' in payload
         f = payload['features']
@@ -100,8 +109,9 @@ class TestBuildHeartbeatPayload:
         assert payload['features']['pipeline_count'] == -1
 
     @pytest.mark.asyncio
-    async def test_cloud_counts_loaded_registries_without_tenant_sql(self):
+    async def test_cloud_counts_loaded_registries_without_tenant_sql(self, monkeypatch):
         heartbeat = get_heartbeat_module()
+        monkeypatch.setattr(heartbeat.constants, 'instance_id', 'instance-test')
         ap = make_app()
         ap.persistence_mgr.mode = SimpleNamespace(value='cloud_runtime')
         ap.persistence_mgr.execute_async = AsyncMock(
@@ -139,8 +149,12 @@ class TestBuildHeartbeatPayload:
         }
         ap.workspace_service.list_active_execution_bindings = AsyncMock(
             return_value=[
-                SimpleNamespace(workspace_uuid='workspace-a'),
-                SimpleNamespace(workspace_uuid='workspace-b'),
+                SimpleNamespace(workspace_uuid='workspace-a', placement_generation=7),
+                SimpleNamespace(
+                    workspace_uuid='workspace-b',
+                    placement_generation=9,
+                    workspace_created_at=datetime(2026, 8, 4, tzinfo=timezone.utc),
+                ),
             ],
         )
         ap.platform_mgr._bots_by_key[('instance-a', 'workspace-b', 'bot-b')] = SimpleNamespace(
@@ -150,7 +164,9 @@ class TestBuildHeartbeatPayload:
         payloads = await heartbeat.build_heartbeat_payloads(ap)
 
         assert [payload['workspace_uuid'] for payload in payloads] == ['workspace-a', 'workspace-b']
-        assert all('instance_id' not in payload for payload in payloads)
+        assert all(payload['instance_id'] for payload in payloads)
+        assert payloads[0]['workspace_create_ts'] == 0
+        assert payloads[1]['workspace_create_ts'] == 1785801600
         by_workspace = {payload['workspace_uuid']: payload['features'] for payload in payloads}
         assert by_workspace['workspace-a']['pipeline_count'] == 2
         assert by_workspace['workspace-a']['mcp_server_count'] == 3
@@ -159,10 +175,12 @@ class TestBuildHeartbeatPayload:
         assert by_workspace['workspace-a']['plugin_count'] == 2
         assert by_workspace['workspace-a']['extension_count'] == 5
         assert by_workspace['workspace-a']['skill_count'] == 2
+        assert by_workspace['workspace-a']['execution_generation'] == 7
         assert by_workspace['workspace-a']['adapters'] == ['WorkspaceAAdapter']
         assert by_workspace['workspace-b']['bot_count'] == 1
         assert by_workspace['workspace-b']['pipeline_count'] == 0
         assert by_workspace['workspace-b']['skill_count'] == 1
+        assert by_workspace['workspace-b']['execution_generation'] == 9
         assert by_workspace['workspace-b']['adapters'] == ['WorkspaceBAdapter']
         assert 'workspace_resources' not in by_workspace['workspace-a']
         ap.persistence_mgr.execute_async.assert_not_awaited()

tests/unit_tests/telemetry/test_telemetry.py

@@ -596,6 +596,36 @@ class TestTelemetryManagedRuntimeAuthentication:
         assert captured['headers'] == {'X-LangBot-Telemetry-Token': 'managed-runtime-secret'}
 
 
+class TestAuthenticatedWorkspaceReporter:
+    @pytest.mark.asyncio
+    async def test_workspace_owner_access_token_is_sent_as_bearer(self):
+        telemetry = get_telemetry_module()
+        mock_app = Mock()
+        mock_app.logger = Mock()
+        mock_app.user_service = Mock()
+        mock_app.user_service.get_workspace_owner = AsyncMock(
+            return_value=Mock(user='owner@example.com', space_access_token='expired-token')
+        )
+        mock_app.space_service = Mock()
+        mock_app.space_service.get_valid_access_token = AsyncMock(return_value='refreshed-workspace-owner-token')
+        manager = telemetry.TelemetryManager(mock_app)
+        manager.telemetry_config = {'url': 'https://example.com'}
+
+        response = Mock(status_code=200, text='')
+        response.json = Mock(return_value={'code': 0})
+        mock_client = Mock()
+        mock_client.post = Mock(return_value=response)
+
+        with patch.object(httpx, 'AsyncClient', return_value=mock_client):
+            await manager.send({'query_id': 'q-1', 'workspace_uuid': 'workspace-1'})
+
+        mock_app.user_service.get_workspace_owner.assert_awaited_once_with('workspace-1')
+        mock_app.space_service.get_valid_access_token.assert_awaited_once_with('owner@example.com')
+        assert mock_client.post.call_args.kwargs['headers'] == {
+            'Authorization': 'Bearer refreshed-workspace-owner-token'
+        }
+
+
 class TestStartSendTask:
     """Tests for start_send_task() method."""
 

tests/unit_tests/telemetry/test_workspace_identity.py

@@ -7,25 +7,28 @@ from types import SimpleNamespace
 def test_standard_oss_instance_id_aligns_to_embedded_uuid():
     from langbot.pkg.workspace.identity import workspace_uuid_from_instance_id
 
-    instance_uuid = "a711d9e4-0953-443f-a0e9-7dd50193a79f"
+    instance_uuid = 'a711d9e4-0953-443f-a0e9-7dd50193a79f'
 
     assert workspace_uuid_from_instance_id(instance_uuid) == instance_uuid
-    assert workspace_uuid_from_instance_id(f"instance_{instance_uuid}") == instance_uuid
+    assert workspace_uuid_from_instance_id(f'instance_{instance_uuid}') == instance_uuid
 
 
 def test_custom_legacy_instance_id_maps_to_stable_valid_uuid():
     from langbot.pkg.workspace.identity import workspace_uuid_from_instance_id
 
-    first = workspace_uuid_from_instance_id("instance_migration_test")
-    second = workspace_uuid_from_instance_id("instance_migration_test")
+    first = workspace_uuid_from_instance_id('instance_migration_test')
+    second = workspace_uuid_from_instance_id('instance_migration_test')
 
     assert first == second
     assert str(uuid.UUID(first)) == first
 
 
-def test_query_telemetry_identity_uses_execution_workspace_only():
+def test_query_telemetry_identity_reports_instance_and_workspace():
     from langbot.pkg.telemetry.identity import workspace_identity
 
-    identity = workspace_identity(SimpleNamespace(workspace_uuid="workspace-a", instance_uuid="instance-a"))
+    identity = workspace_identity(SimpleNamespace(workspace_uuid='workspace-a', instance_uuid='instance-a'))
 
-    assert identity == {"workspace_uuid": "workspace-a"}
+    assert identity == {
+        'instance_id': 'instance-a',
+        'workspace_uuid': 'workspace-a',
+    }

tests/unit_tests/vector/test_mgr.py

@@ -213,7 +213,7 @@ class TestVectorDBManagerInitialization:
                 mock_app,
                 connection_string='postgresql://user:pass@host:5432/langbot',
                 use_business_database=False,
-                allowed_dimensions=[384, 512, 768, 1024, 1536],
+                allowed_dimensions=[384, 512, 768, 1024, 1536, 3072],
             )
 
     def test_initialize_pgvector_with_individual_params(self):
@@ -251,7 +251,7 @@ class TestVectorDBManagerInitialization:
                 user='admin',
                 password='secret',
                 use_business_database=False,
-                allowed_dimensions=[384, 512, 768, 1024, 1536],
+                allowed_dimensions=[384, 512, 768, 1024, 1536, 3072],
             )
 
     def test_initialize_pgvector_defaults(self):
@@ -280,7 +280,7 @@ class TestVectorDBManagerInitialization:
                 user='postgres',
                 password='postgres',
                 use_business_database=False,
-                allowed_dimensions=[384, 512, 768, 1024, 1536],
+                allowed_dimensions=[384, 512, 768, 1024, 1536, 3072],
             )
 
     def test_initialize_pgvector_with_shared_business_database(self):

tests/unit_tests/workspace/test_workspace_collaboration.py

@@ -99,6 +99,7 @@ async def test_invitation_secret_is_hashed_and_acceptance_is_one_time(collaborat
     membership = await service.accept_invitation(created.token, account.uuid)
     assert membership.workspace_uuid == workspace.uuid
     assert membership.role == 'developer'
+    assert membership.source == 'local'
 
     with pytest.raises(InvitationUsedError):
         await service.accept_invitation(created.token, account.uuid)
@@ -203,20 +204,21 @@ async def test_last_owner_cannot_be_demoted(collaboration_context):
             second_membership,
         )
 
-    promoted = await service.update_member_role(
-        workspace.uuid,
-        second.uuid,
-        'owner',
-        owner_membership,
-    )
-    assert promoted.role == 'owner'
-    demoted = await service.update_member_role(
-        workspace.uuid,
-        owner_membership.account_uuid,
-        'admin',
-        owner_membership,
-    )
-    assert demoted.role == 'admin'
+    with pytest.raises(MembershipPermissionError, match='cannot be transferred'):
+        await service.update_member_role(
+            workspace.uuid,
+            second.uuid,
+            'owner',
+            owner_membership,
+        )
+
+    with pytest.raises(LastOwnerError):
+        await service.update_member_role(
+            workspace.uuid,
+            owner_membership.account_uuid,
+            'admin',
+            owner_membership,
+        )
 
 
 async def test_workspace_selector_requires_membership(collaboration_context):

tests/unit_tests/workspace/test_workspace_service.py

@@ -153,6 +153,33 @@ async def test_initial_owner_cannot_be_claimed_by_another_account(workspace_test
         ).all()
         assert len(owners) == 1
         assert owners[0].account_uuid == first_account_uuid
+        assert owners[0].source == 'local'
+
+
+async def test_claim_initial_owner_reclassifies_existing_membership_as_local(workspace_test_context):
+    service, session_factory = workspace_test_context
+
+    async with session_factory() as session:
+        async with session.begin():
+            account_uuid = await _insert_account(session, 'reclaimed@example.com')
+            workspace = await service.ensure_singleton_workspace(session=session)
+            session.add(
+                WorkspaceMembership(
+                    uuid='44444444-4444-4444-8444-444444444444',
+                    workspace_uuid=workspace.uuid,
+                    account_uuid=account_uuid,
+                    role='viewer',
+                    status='removed',
+                    source='cloud_projection',
+                    projection_revision=4,
+                )
+            )
+
+    membership = await service.claim_initial_owner(account_uuid)
+
+    assert membership.role == 'owner'
+    assert membership.status == 'active'
+    assert membership.source == 'local'
 
 
 async def test_execution_binding_returns_persisted_generation(workspace_test_context):

web/.env.example

@@ -1 +1,5 @@
-VITE_API_BASE_URL=http://localhost:5300
+# Leave empty in development to use Vite's same-origin proxy. This keeps API,
+# login, and WebSocket requests working when the UI is opened from another
+# device on the local network.
+VITE_API_BASE_URL=
+VITE_API_PROXY_TARGET=http://127.0.0.1:5300