Merge branch 'rc/new-plugin' into refactor/new-plugin-system

pkg/provider/modelmgr/entities.py

@@ -17,7 +17,7 @@ class LLMModelInfo(pydantic.BaseModel):
 
     token_mgr: token.TokenManager
 
-    requester: requester.LLMAPIRequester
+    requester: requester.ProviderAPIRequester
 
     tool_call_supported: typing.Optional[bool] = False
 

pkg/provider/modelmgr/modelmgr.py

@@ -20,13 +20,16 @@ class ModelManager:
 
     llm_models: list[requester.RuntimeLLMModel]
 
+    embedding_models: list[requester.RuntimeEmbeddingModel]
+
     requester_components: list[engine.Component]
 
-    requester_dict: dict[str, type[requester.LLMAPIRequester]]  # cache
+    requester_dict: dict[str, type[requester.ProviderAPIRequester]]  # cache
 
     def __init__(self, ap: app.Application):
         self.ap = ap
         self.llm_models = []
+        self.embedding_models = []
         self.requester_components = []
         self.requester_dict = {}
 
@@ -34,7 +37,7 @@ class ModelManager:
         self.requester_components = self.ap.discover.get_components_by_kind('LLMAPIRequester')
 
         # forge requester class dict
-        requester_dict: dict[str, type[requester.LLMAPIRequester]] = {}
+        requester_dict: dict[str, type[requester.ProviderAPIRequester]] = {}
         for component in self.requester_components:
             requester_dict[component.metadata.name] = component.get_python_component_class()
 
@@ -47,13 +50,11 @@ class ModelManager:
         self.ap.logger.info('Loading models from db...')
 
         self.llm_models = []
+        self.embedding_models = []
 
         # llm models
         result = await self.ap.persistence_mgr.execute_async(sqlalchemy.select(persistence_model.LLMModel))
-
         llm_models = result.all()
-
-        # load models
         for llm_model in llm_models:
             try:
                 await self.load_llm_model(llm_model)
@@ -62,11 +63,17 @@ class ModelManager:
             except Exception as e:
                 self.ap.logger.error(f'Failed to load model {llm_model.uuid}: {e}\n{traceback.format_exc()}')
 
+        # embedding models
+        result = await self.ap.persistence_mgr.execute_async(sqlalchemy.select(persistence_model.EmbeddingModel))
+        embedding_models = result.all()
+        for embedding_model in embedding_models:
+            await self.load_embedding_model(embedding_model)
+
     async def init_runtime_llm_model(
         self,
         model_info: persistence_model.LLMModel | sqlalchemy.Row[persistence_model.LLMModel] | dict,
     ):
-        """初始化运行时模型"""
+        """初始化运行时 LLM 模型"""
         if isinstance(model_info, sqlalchemy.Row):
             model_info = persistence_model.LLMModel(**model_info._mapping)
         elif isinstance(model_info, dict):
@@ -90,31 +97,85 @@ class ModelManager:
 
         return runtime_llm_model
 
+    async def init_runtime_embedding_model(
+        self,
+        model_info: persistence_model.EmbeddingModel | sqlalchemy.Row[persistence_model.EmbeddingModel] | dict,
+    ):
+        """初始化运行时 Embedding 模型"""
+        if isinstance(model_info, sqlalchemy.Row):
+            model_info = persistence_model.EmbeddingModel(**model_info._mapping)
+        elif isinstance(model_info, dict):
+            model_info = persistence_model.EmbeddingModel(**model_info)
+
+        requester_inst = self.requester_dict[model_info.requester](ap=self.ap, config=model_info.requester_config)
+
+        await requester_inst.initialize()
+
+        runtime_embedding_model = requester.RuntimeEmbeddingModel(
+            model_entity=model_info,
+            token_mgr=token.TokenManager(
+                name=model_info.uuid,
+                tokens=model_info.api_keys,
+            ),
+            requester=requester_inst,
+        )
+
+        return runtime_embedding_model
+
     async def load_llm_model(
         self,
         model_info: persistence_model.LLMModel | sqlalchemy.Row[persistence_model.LLMModel] | dict,
     ):
-        """加载模型"""
+        """加载 LLM 模型"""
         runtime_llm_model = await self.init_runtime_llm_model(model_info)
         self.llm_models.append(runtime_llm_model)
 
+    async def load_embedding_model(
+        self,
+        model_info: persistence_model.EmbeddingModel | sqlalchemy.Row[persistence_model.EmbeddingModel] | dict,
+    ):
+        """加载 Embedding 模型"""
+        runtime_embedding_model = await self.init_runtime_embedding_model(model_info)
+        self.embedding_models.append(runtime_embedding_model)
+
     async def get_model_by_uuid(self, uuid: str) -> requester.RuntimeLLMModel:
-        """通过uuid获取模型"""
+        """通过uuid获取 LLM 模型"""
         for model in self.llm_models:
             if model.model_entity.uuid == uuid:
                 return model
-        raise ValueError(f'model {uuid} not found')
+        raise ValueError(f'LLM model {uuid} not found')
+
+    async def get_embedding_model_by_uuid(self, uuid: str) -> requester.RuntimeEmbeddingModel:
+        """通过uuid获取 Embedding 模型"""
+        for model in self.embedding_models:
+            if model.model_entity.uuid == uuid:
+                return model
+        raise ValueError(f'Embedding model {uuid} not found')
 
     async def remove_llm_model(self, model_uuid: str):
-        """移除模型"""
+        """移除 LLM 模型"""
         for model in self.llm_models:
             if model.model_entity.uuid == model_uuid:
                 self.llm_models.remove(model)
                 return
 
-    def get_available_requesters_info(self) -> list[dict]:
+    async def remove_embedding_model(self, model_uuid: str):
+        """移除 Embedding 模型"""
+        for model in self.embedding_models:
+            if model.model_entity.uuid == model_uuid:
+                self.embedding_models.remove(model)
+                return
+
+    def get_available_requesters_info(self, model_type: str) -> list[dict]:
         """获取所有可用的请求器"""
-        return [component.to_plain_dict() for component in self.requester_components]
+        if model_type != '':
+            return [
+                component.to_plain_dict()
+                for component in self.requester_components
+                if model_type in component.spec['support_type']
+            ]
+        else:
+            return [component.to_plain_dict() for component in self.requester_components]
 
     def get_available_requester_info_by_name(self, name: str) -> dict | None:
         """通过名称获取请求器信息"""

pkg/provider/modelmgr/requester.py

@@ -20,22 +20,45 @@ class RuntimeLLMModel:
     token_mgr: token.TokenManager
     """api key管理器"""
 
-    requester: LLMAPIRequester
+    requester: ProviderAPIRequester
     """请求器实例"""
 
     def __init__(
         self,
         model_entity: persistence_model.LLMModel,
         token_mgr: token.TokenManager,
-        requester: LLMAPIRequester,
+        requester: ProviderAPIRequester,
     ):
         self.model_entity = model_entity
         self.token_mgr = token_mgr
         self.requester = requester
 
 
-class LLMAPIRequester(metaclass=abc.ABCMeta):
-    """LLM API请求器"""
+class RuntimeEmbeddingModel:
+    """运行时 Embedding 模型"""
+
+    model_entity: persistence_model.EmbeddingModel
+    """模型数据"""
+
+    token_mgr: token.TokenManager
+    """api key管理器"""
+
+    requester: ProviderAPIRequester
+    """请求器实例"""
+
+    def __init__(
+        self,
+        model_entity: persistence_model.EmbeddingModel,
+        token_mgr: token.TokenManager,
+        requester: ProviderAPIRequester,
+    ):
+        self.model_entity = model_entity
+        self.token_mgr = token_mgr
+        self.requester = requester
+
+
+class ProviderAPIRequester(metaclass=abc.ABCMeta):
+    """Provider API请求器"""
 
     name: str = None
 
@@ -61,6 +84,7 @@ class LLMAPIRequester(metaclass=abc.ABCMeta):
         messages: typing.List[provider_message.Message],
         funcs: typing.List[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         """调用API
 
@@ -69,8 +93,50 @@ class LLMAPIRequester(metaclass=abc.ABCMeta):
             messages (typing.List[llm_entities.Message]): 消息对象列表
             funcs (typing.List[tools_entities.LLMFunction], optional): 使用的工具函数列表. Defaults to None.
             extra_args (dict[str, typing.Any], optional): 额外的参数. Defaults to {}.
+            remove_think (bool, optional): 是否移思考中的消息. Defaults to False.
 
         Returns:
             llm_entities.Message: 返回消息对象
         """
         pass
+
+    async def invoke_llm_stream(
+        self,
+        query: pipeline_query.Query,
+        model: RuntimeLLMModel,
+        messages: typing.List[provider_message.Message],
+        funcs: typing.List[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.MessageChunk:
+        """调用API
+
+        Args:
+            model (RuntimeLLMModel): 使用的模型信息
+            messages (typing.List[provider_message.Message]): 消息对象列表
+            funcs (typing.List[resource_tool.LLMTool], optional): 使用的工具函数列表. Defaults to None.
+            extra_args (dict[str, typing.Any], optional): 额外的参数. Defaults to {}.
+            remove_think (bool, optional): 是否移除思考中的消息. Defaults to False.
+
+        Returns:
+            typing.AsyncGenerator[provider_message.MessageChunk]: 返回消息对象
+        """
+        pass
+
+    async def invoke_embedding(
+        self,
+        model: RuntimeEmbeddingModel,
+        input_text: typing.List[str],
+        extra_args: dict[str, typing.Any] = {},
+    ) -> typing.List[typing.List[float]]:
+        """调用 Embedding API
+
+        Args:
+            model (RuntimeEmbeddingModel): 使用的模型信息
+            input_text (typing.List[str]): 输入文本
+            extra_args (dict[str, typing.Any], optional): 额外的参数. Defaults to {}.
+
+        Returns:
+            typing.List[typing.List[float]]: 返回的 embedding 向量
+        """
+        pass

pkg/provider/modelmgr/requesters/302aichatcmpl.py

@@ -7,7 +7,7 @@ from . import chatcmpl
 
 
 class AI302ChatCompletions(chatcmpl.OpenAIChatCompletions):
-    """302 AI ChatCompletion API 请求器"""
+    """302.AI ChatCompletion API 请求器"""
 
     client: openai.AsyncClient
 

pkg/provider/modelmgr/requesters/302aichatcmpl.yaml

@@ -3,8 +3,8 @@ kind: LLMAPIRequester
 metadata:
   name: 302-ai-chat-completions
   label:
-    en_US: 302 AI
-    zh_Hans: 302 AI
+    en_US: 302.AI
+    zh_Hans: 302.AI
   icon: 302ai.png
 spec:
   config:
@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./302aichatcmpl.py

pkg/provider/modelmgr/requesters/anthropicmsgs.py

@@ -15,13 +15,13 @@ import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
 
 
-class AnthropicMessages(requester.LLMAPIRequester):
+class AnthropicMessages(requester.ProviderAPIRequester):
     """Anthropic Messages API 请求器"""
 
     client: anthropic.AsyncAnthropic
 
     default_config: dict[str, typing.Any] = {
-        'base_url': 'https://api.anthropic.com/v1',
+        'base_url': 'https://api.anthropic.com',
         'timeout': 120,
     }
 
@@ -44,6 +44,7 @@ class AnthropicMessages(requester.LLMAPIRequester):
         self.client = anthropic.AsyncAnthropic(
             api_key='',
             http_client=httpx_client,
+            base_url=self.requester_cfg['base_url'],
         )
 
     async def invoke_llm(
@@ -53,6 +54,7 @@ class AnthropicMessages(requester.LLMAPIRequester):
         messages: typing.List[provider_message.Message],
         funcs: typing.List[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         self.client.api_key = model.token_mgr.get_token()
 
@@ -89,7 +91,8 @@ class AnthropicMessages(requester.LLMAPIRequester):
                             {
                                 'type': 'tool_result',
                                 'tool_use_id': tool_call_id,
-                                'content': m.content,
+                                'is_error': False,
+                                'content': [{'type': 'text', 'text': m.content}],
                             }
                         ],
                     }
@@ -133,6 +136,9 @@ class AnthropicMessages(requester.LLMAPIRequester):
 
         args['messages'] = req_messages
 
+        if 'thinking' in args:
+            args['thinking'] = {'type': 'enabled', 'budget_tokens': 10000}
+
         if funcs:
             tools = await self.ap.tool_mgr.generate_tools_for_anthropic(funcs)
 
@@ -140,19 +146,17 @@ class AnthropicMessages(requester.LLMAPIRequester):
                 args['tools'] = tools
 
         try:
-            # print(json.dumps(args, indent=4, ensure_ascii=False))
             resp = await self.client.messages.create(**args)
 
             args = {
                 'content': '',
                 'role': resp.role,
             }
-
             assert type(resp) is anthropic.types.message.Message
 
             for block in resp.content:
-                if block.type == 'thinking':
-                    args['content'] = '<think>' + block.thinking + '</think>\n' + args['content']
+                if not remove_think and block.type == 'thinking':
+                    args['content'] = '<think>\n' + block.thinking + '\n</think>\n' + args['content']
                 elif block.type == 'text':
                     args['content'] += block.text
                 elif block.type == 'tool_use':
@@ -176,3 +180,191 @@ class AnthropicMessages(requester.LLMAPIRequester):
                 raise errors.RequesterError(f'模型无效: {e.message}')
             else:
                 raise errors.RequesterError(f'请求地址无效: {e.message}')
+
+    async def invoke_llm_stream(
+        self,
+        query: pipeline_query.Query,
+        model: requester.RuntimeLLMModel,
+        messages: typing.List[provider_message.Message],
+        funcs: typing.List[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.Message:
+        self.client.api_key = model.token_mgr.get_token()
+
+        args = extra_args.copy()
+        args['model'] = model.model_entity.name
+        args['stream'] = True
+
+        # 处理消息
+
+        # system
+        system_role_message = None
+
+        for i, m in enumerate(messages):
+            if m.role == 'system':
+                system_role_message = m
+
+                break
+
+        if system_role_message:
+            messages.pop(i)
+
+        if isinstance(system_role_message, provider_message.Message) and isinstance(system_role_message.content, str):
+            args['system'] = system_role_message.content
+
+        req_messages = []
+
+        for m in messages:
+            if m.role == 'tool':
+                tool_call_id = m.tool_call_id
+
+                req_messages.append(
+                    {
+                        'role': 'user',
+                        'content': [
+                            {
+                                'type': 'tool_result',
+                                'tool_use_id': tool_call_id,
+                                'is_error': False,  # 暂时直接写false
+                                'content': [
+                                    {'type': 'text', 'text': m.content}
+                                ],  # 这里要是list包裹，应该是多个返回的情况？type类型好像也可以填其他的，暂时只写text
+                            }
+                        ],
+                    }
+                )
+
+                continue
+
+            msg_dict = m.dict(exclude_none=True)
+
+            if isinstance(m.content, str) and m.content.strip() != '':
+                msg_dict['content'] = [{'type': 'text', 'text': m.content}]
+            elif isinstance(m.content, list):
+                for i, ce in enumerate(m.content):
+                    if ce.type == 'image_base64':
+                        image_b64, image_format = await image.extract_b64_and_format(ce.image_base64)
+
+                        alter_image_ele = {
+                            'type': 'image',
+                            'source': {
+                                'type': 'base64',
+                                'media_type': f'image/{image_format}',
+                                'data': image_b64,
+                            },
+                        }
+                        msg_dict['content'][i] = alter_image_ele
+            if isinstance(msg_dict['content'], str) and msg_dict['content'] == '':
+                msg_dict['content'] = []  # 这里不知道为什么会莫名有个空导致content为字符
+            if m.tool_calls:
+                for tool_call in m.tool_calls:
+                    msg_dict['content'].append(
+                        {
+                            'type': 'tool_use',
+                            'id': tool_call.id,
+                            'name': tool_call.function.name,
+                            'input': json.loads(tool_call.function.arguments),
+                        }
+                    )
+
+                del msg_dict['tool_calls']
+
+            req_messages.append(msg_dict)
+        if 'thinking' in args:
+            args['thinking'] = {'type': 'enabled', 'budget_tokens': 10000}
+
+        args['messages'] = req_messages
+
+        if funcs:
+            tools = await self.ap.tool_mgr.generate_tools_for_anthropic(funcs)
+
+            if tools:
+                args['tools'] = tools
+
+        try:
+            role = 'assistant'  # 默认角色
+            # chunk_idx = 0
+            think_started = False
+            think_ended = False
+            finish_reason = False
+            content = ''
+            tool_name = ''
+            tool_id = ''
+            async for chunk in await self.client.messages.create(**args):
+                tool_call = {'id': None, 'function': {'name': None, 'arguments': None}, 'type': 'function'}
+                if isinstance(
+                    chunk, anthropic.types.raw_content_block_start_event.RawContentBlockStartEvent
+                ):  # 记录开始
+                    if chunk.content_block.type == 'tool_use':
+                        if chunk.content_block.name is not None:
+                            tool_name = chunk.content_block.name
+                        if chunk.content_block.id is not None:
+                            tool_id = chunk.content_block.id
+
+                        tool_call['function']['name'] = tool_name
+                        tool_call['function']['arguments'] = ''
+                        tool_call['id'] = tool_id
+
+                    if not remove_think:
+                        if chunk.content_block.type == 'thinking' and not remove_think:
+                            think_started = True
+                        elif chunk.content_block.type == 'text' and chunk.index != 0 and not remove_think:
+                            think_ended = True
+                        continue
+                elif isinstance(chunk, anthropic.types.raw_content_block_delta_event.RawContentBlockDeltaEvent):
+                    if chunk.delta.type == 'thinking_delta':
+                        if think_started:
+                            think_started = False
+                            content = '<think>\n' + chunk.delta.thinking
+                        elif remove_think:
+                            continue
+                        else:
+                            content = chunk.delta.thinking
+                    elif chunk.delta.type == 'text_delta':
+                        if think_ended:
+                            think_ended = False
+                            content = '\n</think>\n' + chunk.delta.text
+                        else:
+                            content = chunk.delta.text
+                    elif chunk.delta.type == 'input_json_delta':
+                        tool_call['function']['arguments'] = chunk.delta.partial_json
+                        tool_call['function']['name'] = tool_name
+                        tool_call['id'] = tool_id
+                elif isinstance(chunk, anthropic.types.raw_content_block_stop_event.RawContentBlockStopEvent):
+                    continue  # 记录raw_content_block结束的
+
+                elif isinstance(chunk, anthropic.types.raw_message_delta_event.RawMessageDeltaEvent):
+                    if chunk.delta.stop_reason == 'end_turn':
+                        finish_reason = True
+                elif isinstance(chunk, anthropic.types.raw_message_stop_event.RawMessageStopEvent):
+                    continue  # 这个好像是完全结束
+                else:
+                    # print(chunk)
+                    self.ap.logger.debug(f'anthropic chunk: {chunk}')
+                    continue
+
+                args = {
+                    'content': content,
+                    'role': role,
+                    'is_final': finish_reason,
+                    'tool_calls': None if tool_call['id'] is None else [tool_call],
+                }
+                # if chunk_idx == 0:
+                #     chunk_idx += 1
+                #     continue
+
+                # assert type(chunk) is anthropic.types.message.Chunk
+
+                yield provider_message.MessageChunk(**args)
+
+            # return llm_entities.Message(**args)
+        except anthropic.AuthenticationError as e:
+            raise errors.RequesterError(f'api-key 无效: {e.message}')
+        except anthropic.BadRequestError as e:
+            raise errors.RequesterError(str(e.message))
+        except anthropic.NotFoundError as e:
+            if 'model: ' in str(e):
+                raise errors.RequesterError(f'模型无效: {e.message}')
+            else:
+                raise errors.RequesterError(f'请求地址无效: {e.message}')

pkg/provider/modelmgr/requesters/anthropicmsgs.yaml

@@ -14,7 +14,7 @@ spec:
         zh_Hans: 基础 URL
       type: string
       required: true
-      default: "https://api.anthropic.com/v1"
+      default: "https://api.anthropic.com"
     - name: timeout
       label:
         en_US: Timeout
@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./anthropicmsgs.py

pkg/provider/modelmgr/requesters/bailianchatcmpl.yaml

@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./bailianchatcmpl.py

pkg/provider/modelmgr/requesters/chatcmpl.py

@@ -13,7 +13,7 @@ import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
 
 
-class OpenAIChatCompletions(requester.LLMAPIRequester):
+class OpenAIChatCompletions(requester.ProviderAPIRequester):
     """OpenAI ChatCompletion API 请求器"""
 
     client: openai.AsyncClient
@@ -38,9 +38,18 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
     ) -> chat_completion.ChatCompletion:
         return await self.client.chat.completions.create(**args, extra_body=extra_body)
 
+    async def _req_stream(
+        self,
+        args: dict,
+        extra_body: dict = {},
+    ):
+        async for chunk in await self.client.chat.completions.create(**args, extra_body=extra_body):
+            yield chunk
+
     async def _make_msg(
         self,
         chat_completion: chat_completion.ChatCompletion,
+        remove_think: bool = False,
     ) -> provider_message.Message:
         chatcmpl_message = chat_completion.choices[0].message.model_dump()
 
@@ -48,16 +57,191 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
         if 'role' not in chatcmpl_message or chatcmpl_message['role'] is None:
             chatcmpl_message['role'] = 'assistant'
 
-        reasoning_content = chatcmpl_message['reasoning_content'] if 'reasoning_content' in chatcmpl_message else None
+        # 处理思维链
+        content = chatcmpl_message.get('content', '')
+        reasoning_content = chatcmpl_message.get('reasoning_content', None)
 
-        # deepseek的reasoner模型
-        if reasoning_content is not None:
-            chatcmpl_message['content'] = '<think>\n' + reasoning_content + '\n</think>\n' + chatcmpl_message['content']
+        processed_content, _ = await self._process_thinking_content(
+            content=content, reasoning_content=reasoning_content, remove_think=remove_think
+        )
+
+        chatcmpl_message['content'] = processed_content
+
+        # 移除 reasoning_content 字段，避免传递给 Message
+        if 'reasoning_content' in chatcmpl_message:
+            del chatcmpl_message['reasoning_content']
 
         message = provider_message.Message(**chatcmpl_message)
 
         return message
 
+    async def _process_thinking_content(
+        self,
+        content: str,
+        reasoning_content: str = None,
+        remove_think: bool = False,
+    ) -> tuple[str, str]:
+        """处理思维链内容
+
+        Args:
+            content: 原始内容
+            reasoning_content: reasoning_content 字段内容
+            remove_think: 是否移除思维链
+
+        Returns:
+            (处理后的内容, 提取的思维链内容)
+        """
+        thinking_content = ''
+
+        # 1. 从 reasoning_content 提取思维链
+        if reasoning_content:
+            thinking_content = reasoning_content
+
+        # 2. 从 content 中提取 <think> 标签内容
+        if content and '<think>' in content and '</think>' in content:
+            import re
+
+            think_pattern = r'<think>(.*?)</think>'
+            think_matches = re.findall(think_pattern, content, re.DOTALL)
+            if think_matches:
+                # 如果已有 reasoning_content，则追加
+                if thinking_content:
+                    thinking_content += '\n' + '\n'.join(think_matches)
+                else:
+                    thinking_content = '\n'.join(think_matches)
+                # 移除 content 中的 <think> 标签
+                content = re.sub(think_pattern, '', content, flags=re.DOTALL).strip()
+
+        # 3. 根据 remove_think 参数决定是否保留思维链
+        if remove_think:
+            return content, ''
+        else:
+            # 如果有思维链内容，将其以 <think> 格式添加到 content 开头
+            if thinking_content:
+                content = f'<think>\n{thinking_content}\n</think>\n{content}'.strip()
+            return content, thinking_content
+
+    async def _closure_stream(
+        self,
+        query: pipeline_query.Query,
+        req_messages: list[dict],
+        use_model: requester.RuntimeLLMModel,
+        use_funcs: list[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.MessageChunk:
+        self.client.api_key = use_model.token_mgr.get_token()
+
+        args = {}
+        args['model'] = use_model.model_entity.name
+
+        if use_funcs:
+            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
+            if tools:
+                args['tools'] = tools
+
+        # 设置此次请求中的messages
+        messages = req_messages.copy()
+
+        # 检查vision
+        for msg in messages:
+            if 'content' in msg and isinstance(msg['content'], list):
+                for me in msg['content']:
+                    if me['type'] == 'image_base64':
+                        me['image_url'] = {'url': me['image_base64']}
+                        me['type'] = 'image_url'
+                        del me['image_base64']
+
+        args['messages'] = messages
+        args['stream'] = True
+
+        # 流式处理状态
+        # tool_calls_map: dict[str, provider_message.ToolCall] = {}
+        chunk_idx = 0
+        thinking_started = False
+        thinking_ended = False
+        role = 'assistant'  # 默认角色
+        tool_id = ''
+        tool_name = ''
+        # accumulated_reasoning = ''  # 仅用于判断何时结束思维链
+
+        async for chunk in self._req_stream(args, extra_body=extra_args):
+            # 解析 chunk 数据
+
+            if hasattr(chunk, 'choices') and chunk.choices:
+                choice = chunk.choices[0]
+                delta = choice.delta.model_dump() if hasattr(choice, 'delta') else {}
+
+                finish_reason = getattr(choice, 'finish_reason', None)
+            else:
+                delta = {}
+                finish_reason = None
+            # 从第一个 chunk 获取 role，后续使用这个 role
+            if 'role' in delta and delta['role']:
+                role = delta['role']
+
+            # 获取增量内容
+            delta_content = delta.get('content', '')
+            reasoning_content = delta.get('reasoning_content', '')
+
+            # 处理 reasoning_content
+            if reasoning_content:
+                # accumulated_reasoning += reasoning_content
+                # 如果设置了 remove_think，跳过 reasoning_content
+                if remove_think:
+                    chunk_idx += 1
+                    continue
+
+                # 第一次出现 reasoning_content，添加 <think> 开始标签
+                if not thinking_started:
+                    thinking_started = True
+                    delta_content = '<think>\n' + reasoning_content
+                else:
+                    # 继续输出 reasoning_content
+                    delta_content = reasoning_content
+            elif thinking_started and not thinking_ended and delta_content:
+                # reasoning_content 结束，normal content 开始，添加 </think> 结束标签
+                thinking_ended = True
+                delta_content = '\n</think>\n' + delta_content
+
+            # 处理 content 中已有的 <think> 标签（如果需要移除）
+            # if delta_content and remove_think and '<think>' in delta_content:
+            #     import re
+            #
+            #     # 移除 <think> 标签及其内容
+            #     delta_content = re.sub(r'<think>.*?</think>', '', delta_content, flags=re.DOTALL)
+
+            # 处理工具调用增量
+            # delta_tool_calls = None
+            if delta.get('tool_calls'):
+                for tool_call in delta['tool_calls']:
+                    if tool_call['id'] and tool_call['function']['name']:
+                        tool_id = tool_call['id']
+                        tool_name = tool_call['function']['name']
+                    else:
+                        tool_call['id'] = tool_id
+                        tool_call['function']['name'] = tool_name
+                    if tool_call['type'] is None:
+                        tool_call['type'] = 'function'
+
+            # 跳过空的第一个 chunk（只有 role 没有内容）
+            if chunk_idx == 0 and not delta_content and not reasoning_content and not delta.get('tool_calls'):
+                chunk_idx += 1
+                continue
+            # 构建 MessageChunk - 只包含增量内容
+            chunk_data = {
+                'role': role,
+                'content': delta_content if delta_content else None,
+                'tool_calls': delta.get('tool_calls'),
+                'is_final': bool(finish_reason),
+            }
+
+            # 移除 None 值
+            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
+
     async def _closure(
         self,
         query: pipeline_query.Query,
@@ -65,6 +249,7 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
         use_model: requester.RuntimeLLMModel,
         use_funcs: list[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         self.client.api_key = use_model.token_mgr.get_token()
 
@@ -92,10 +277,10 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
         args['messages'] = messages
 
         # 发送请求
-        resp = await self._req(args, extra_body=extra_args)
 
+        resp = await self._req(args, extra_body=extra_args)
         # 处理请求结果
-        message = await self._make_msg(resp)
+        message = await self._make_msg(resp, remove_think)
 
         return message
 
@@ -106,6 +291,7 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
         messages: typing.List[provider_message.Message],
         funcs: typing.List[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         req_messages = []  # req_messages 仅用于类内，外部同步由 query.messages 进行
         for m in messages:
@@ -119,13 +305,90 @@ class OpenAIChatCompletions(requester.LLMAPIRequester):
             req_messages.append(msg_dict)
 
         try:
-            return await self._closure(
+            msg = await self._closure(
                 query=query,
                 req_messages=req_messages,
                 use_model=model,
                 use_funcs=funcs,
                 extra_args=extra_args,
+                remove_think=remove_think,
             )
+            return msg
+        except asyncio.TimeoutError:
+            raise errors.RequesterError('请求超时')
+        except openai.BadRequestError as e:
+            if 'context_length_exceeded' in e.message:
+                raise errors.RequesterError(f'上文过长，请重置会话: {e.message}')
+            else:
+                raise errors.RequesterError(f'请求参数错误: {e.message}')
+        except openai.AuthenticationError as e:
+            raise errors.RequesterError(f'无效的 api-key: {e.message}')
+        except openai.NotFoundError as e:
+            raise errors.RequesterError(f'请求路径错误: {e.message}')
+        except openai.RateLimitError as e:
+            raise errors.RequesterError(f'请求过于频繁或余额不足: {e.message}')
+        except openai.APIError as e:
+            raise errors.RequesterError(f'请求错误: {e.message}')
+
+    async def invoke_embedding(
+        self,
+        model: requester.RuntimeEmbeddingModel,
+        input_text: list[str],
+        extra_args: dict[str, typing.Any] = {},
+    ) -> list[list[float]]:
+        """调用 Embedding API"""
+        self.client.api_key = model.token_mgr.get_token()
+
+        args = {
+            'model': model.model_entity.name,
+            'input': input_text,
+        }
+
+        if model.model_entity.extra_args:
+            args.update(model.model_entity.extra_args)
+
+        args.update(extra_args)
+
+        try:
+            resp = await self.client.embeddings.create(**args)
+
+            return [d.embedding for d in resp.data]
+        except asyncio.TimeoutError:
+            raise errors.RequesterError('请求超时')
+        except openai.BadRequestError as e:
+            raise errors.RequesterError(f'请求参数错误: {e.message}')
+
+    async def invoke_llm_stream(
+        self,
+        query: pipeline_query.Query,
+        model: requester.RuntimeLLMModel,
+        messages: typing.List[provider_message.Message],
+        funcs: typing.List[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.MessageChunk:
+        req_messages = []  # req_messages 仅用于类内，外部同步由 query.messages 进行
+        for m in messages:
+            msg_dict = m.dict(exclude_none=True)
+            content = msg_dict.get('content')
+            if isinstance(content, list):
+                # 检查 content 列表中是否每个部分都是文本
+                if all(isinstance(part, dict) and part.get('type') == 'text' for part in content):
+                    # 将所有文本部分合并为一个字符串
+                    msg_dict['content'] = '\n'.join(part['text'] for part in content)
+            req_messages.append(msg_dict)
+
+        try:
+            async for item in self._closure_stream(
+                query=query,
+                req_messages=req_messages,
+                use_model=model,
+                use_funcs=funcs,
+                extra_args=extra_args,
+                remove_think=remove_think,
+            ):
+                yield item
+
         except asyncio.TimeoutError:
             raise errors.RequesterError('请求超时')
         except openai.BadRequestError as e:

pkg/provider/modelmgr/requesters/chatcmpl.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./chatcmpl.py

pkg/provider/modelmgr/requesters/compsharechatcmpl.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+import typing
+import openai
+
+from . import chatcmpl
+
+
+class CompShareChatCompletions(chatcmpl.OpenAIChatCompletions):
+    """CompShare ChatCompletion API 请求器"""
+
+    client: openai.AsyncClient
+
+    default_config: dict[str, typing.Any] = {
+        'base_url': 'https://api.modelverse.cn/v1',
+        'timeout': 120,
+    }

pkg/provider/modelmgr/requesters/compsharechatcmpl.yaml

@@ -0,0 +1,30 @@
+apiVersion: v1
+kind: LLMAPIRequester
+metadata:
+  name: compshare-chat-completions
+  label:
+    en_US: CompShare
+    zh_Hans: 优云智算
+  icon: compshare.png
+spec:
+  config:
+    - name: base_url
+      label:
+        en_US: Base URL
+        zh_Hans: 基础 URL
+      type: string
+      required: true
+      default: "https://api.modelverse.cn/v1"
+    - name: timeout
+      label:
+        en_US: Timeout
+        zh_Hans: 超时时间
+      type: integer
+      required: true
+      default: 120
+  support_type:
+    - llm
+execution:
+  python:
+    path: ./compsharechatcmpl.py
+    attr: CompShareChatCompletions

pkg/provider/modelmgr/requesters/deepseekchatcmpl.py

@@ -24,6 +24,7 @@ class DeepseekChatCompletions(chatcmpl.OpenAIChatCompletions):
         use_model: requester.RuntimeLLMModel,
         use_funcs: list[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         self.client.api_key = use_model.token_mgr.get_token()
 
@@ -49,10 +50,11 @@ class DeepseekChatCompletions(chatcmpl.OpenAIChatCompletions):
         # 发送请求
         resp = await self._req(args, extra_body=extra_args)
 
+        # print(resp)
+
         if resp is None:
             raise errors.RequesterError('接口返回为空，请确定模型提供商服务是否正常')
-
         # 处理请求结果
-        message = await self._make_msg(resp)
+        message = await self._make_msg(resp, remove_think)
 
         return message

pkg/provider/modelmgr/requesters/deepseekchatcmpl.yaml

@@ -4,7 +4,7 @@ metadata:
   name: deepseek-chat-completions
   label:
     en_US: DeepSeek
-    zh_Hans: 深度求索
+    zh_Hans: DeepSeek
   icon: deepseek.svg
 spec:
   config:
@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./deepseekchatcmpl.py

pkg/provider/modelmgr/requesters/geminichatcmpl.py

@@ -4,6 +4,13 @@ import typing
 
 from . import chatcmpl
 
+import uuid
+
+from .. import requester
+import langbot_plugin.api.entities.builtin.provider.message as provider_message
+import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
+import langbot_plugin.api.entities.builtin.resource.tool as resource_tool
+
 
 class GeminiChatCompletions(chatcmpl.OpenAIChatCompletions):
     """Google Gemini API 请求器"""
@@ -12,3 +19,124 @@ class GeminiChatCompletions(chatcmpl.OpenAIChatCompletions):
         'base_url': 'https://generativelanguage.googleapis.com/v1beta/openai',
         'timeout': 120,
     }
+
+    async def _closure_stream(
+        self,
+        query: pipeline_query.Query,
+        req_messages: list[dict],
+        use_model: requester.RuntimeLLMModel,
+        use_funcs: list[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.MessageChunk:
+        self.client.api_key = use_model.token_mgr.get_token()
+
+        args = {}
+        args['model'] = use_model.model_entity.name
+
+        if use_funcs:
+            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
+            if tools:
+                args['tools'] = tools
+
+        # 设置此次请求中的messages
+        messages = req_messages.copy()
+
+        # 检查vision
+        for msg in messages:
+            if 'content' in msg and isinstance(msg['content'], list):
+                for me in msg['content']:
+                    if me['type'] == 'image_base64':
+                        me['image_url'] = {'url': me['image_base64']}
+                        me['type'] = 'image_url'
+                        del me['image_base64']
+
+        args['messages'] = messages
+        args['stream'] = True
+
+        # 流式处理状态
+        # tool_calls_map: dict[str, provider_message.ToolCall] = {}
+        chunk_idx = 0
+        thinking_started = False
+        thinking_ended = False
+        role = 'assistant'  # 默认角色
+        tool_id = ''
+        tool_name = ''
+        # accumulated_reasoning = ''  # 仅用于判断何时结束思维链
+
+        async for chunk in self._req_stream(args, extra_body=extra_args):
+            # 解析 chunk 数据
+
+            if hasattr(chunk, 'choices') and chunk.choices:
+                choice = chunk.choices[0]
+                delta = choice.delta.model_dump() if hasattr(choice, 'delta') else {}
+
+                finish_reason = getattr(choice, 'finish_reason', None)
+            else:
+                delta = {}
+                finish_reason = None
+            # 从第一个 chunk 获取 role，后续使用这个 role
+            if 'role' in delta and delta['role']:
+                role = delta['role']
+
+            # 获取增量内容
+            delta_content = delta.get('content', '')
+            reasoning_content = delta.get('reasoning_content', '')
+
+            # 处理 reasoning_content
+            if reasoning_content:
+                # accumulated_reasoning += reasoning_content
+                # 如果设置了 remove_think，跳过 reasoning_content
+                if remove_think:
+                    chunk_idx += 1
+                    continue
+
+                # 第一次出现 reasoning_content，添加 <think> 开始标签
+                if not thinking_started:
+                    thinking_started = True
+                    delta_content = '<think>\n' + reasoning_content
+                else:
+                    # 继续输出 reasoning_content
+                    delta_content = reasoning_content
+            elif thinking_started and not thinking_ended and delta_content:
+                # reasoning_content 结束，normal content 开始，添加 </think> 结束标签
+                thinking_ended = True
+                delta_content = '\n</think>\n' + delta_content
+
+            # 处理 content 中已有的 <think> 标签（如果需要移除）
+            # if delta_content and remove_think and '<think>' in delta_content:
+            #     import re
+            #
+            #     # 移除 <think> 标签及其内容
+            #     delta_content = re.sub(r'<think>.*?</think>', '', delta_content, flags=re.DOTALL)
+
+            # 处理工具调用增量
+            # delta_tool_calls = None
+            if delta.get('tool_calls'):
+                for tool_call in delta['tool_calls']:
+                    if tool_call['id'] == '' and tool_id == '':
+                        tool_id = str(uuid.uuid4())
+                    if tool_call['function']['name']:
+                        tool_name = tool_call['function']['name']
+                    tool_call['id'] = tool_id
+                    tool_call['function']['name'] = tool_name
+                    if tool_call['type'] is None:
+                        tool_call['type'] = 'function'
+
+            # 跳过空的第一个 chunk（只有 role 没有内容）
+            if chunk_idx == 0 and not delta_content and not reasoning_content and not delta.get('tool_calls'):
+                chunk_idx += 1
+                continue
+            # 构建 MessageChunk - 只包含增量内容
+            chunk_data = {
+                'role': role,
+                'content': delta_content if delta_content else None,
+                'tool_calls': delta.get('tool_calls'),
+                'is_final': bool(finish_reason),
+            }
+
+            # 移除 None 值
+            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

pkg/provider/modelmgr/requesters/geminichatcmpl.yaml

@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./geminichatcmpl.py

pkg/provider/modelmgr/requesters/giteeaichatcmpl.py

@@ -3,49 +3,13 @@ from __future__ import annotations
 
 import typing
 
-from . import chatcmpl
-from .. import requester
-import langbot_plugin.api.entities.builtin.resource.tool as resource_tool
-import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
-import langbot_plugin.api.entities.builtin.provider.message as provider_message
+from . import ppiochatcmpl
 
 
-class GiteeAIChatCompletions(chatcmpl.OpenAIChatCompletions):
+class GiteeAIChatCompletions(ppiochatcmpl.PPIOChatCompletions):
     """Gitee AI ChatCompletions API 请求器"""
 
     default_config: dict[str, typing.Any] = {
         'base_url': 'https://ai.gitee.com/v1',
         'timeout': 120,
     }
-
-    async def _closure(
-        self,
-        query: pipeline_query.Query,
-        req_messages: list[dict],
-        use_model: requester.RuntimeLLMModel,
-        use_funcs: list[resource_tool.LLMTool] = None,
-        extra_args: dict[str, typing.Any] = {},
-    ) -> provider_message.Message:
-        self.client.api_key = use_model.token_mgr.get_token()
-
-        args = {}
-        args['model'] = use_model.model_entity.name
-
-        if use_funcs:
-            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
-
-            if tools:
-                args['tools'] = tools
-
-        # gitee 不支持多模态，把content都转换成纯文字
-        for m in req_messages:
-            if 'content' in m and isinstance(m['content'], list):
-                m['content'] = ' '.join([c['text'] for c in m['content']])
-
-        args['messages'] = req_messages
-
-        resp = await self._req(args, extra_body=extra_args)
-
-        message = await self._make_msg(resp)
-
-        return message

pkg/provider/modelmgr/requesters/giteeaichatcmpl.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./giteeaichatcmpl.py

pkg/provider/modelmgr/requesters/lmstudiochatcmpl.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./lmstudiochatcmpl.py

pkg/provider/modelmgr/requesters/modelscopechatcmpl.py

@@ -5,7 +5,6 @@ import typing
 
 import openai
 import openai.types.chat.chat_completion as chat_completion
-import openai.types.chat.chat_completion_message_tool_call as chat_completion_message_tool_call
 import httpx
 
 from .. import entities, errors, requester
@@ -14,7 +13,7 @@ import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
 
 
-class ModelScopeChatCompletions(requester.LLMAPIRequester):
+class ModelScopeChatCompletions(requester.ProviderAPIRequester):
     """ModelScope ChatCompletion API 请求器"""
 
     client: openai.AsyncClient
@@ -34,9 +33,11 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
 
     async def _req(
         self,
+        query: pipeline_query.Query,
         args: dict,
         extra_body: dict = {},
-    ) -> chat_completion.ChatCompletion:
+        remove_think: bool = False,
+    ) -> list[dict[str, typing.Any]]:
         args['stream'] = True
 
         chunk = None
@@ -47,73 +48,71 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
 
         resp_gen: openai.AsyncStream = await self.client.chat.completions.create(**args, extra_body=extra_body)
 
+        chunk_idx = 0
+        thinking_started = False
+        thinking_ended = False
+        tool_id = ''
+        tool_name = ''
+        message_delta = {}
         async for chunk in resp_gen:
-            # print(chunk)
             if not chunk or not chunk.id or not chunk.choices or not chunk.choices[0] or not chunk.choices[0].delta:
                 continue
 
-            if chunk.choices[0].delta.content is not None:
-                pending_content += chunk.choices[0].delta.content
+            delta = chunk.choices[0].delta.model_dump() if hasattr(chunk.choices[0], 'delta') else {}
+            reasoning_content = delta.get('reasoning_content')
+            # 处理 reasoning_content
+            if reasoning_content:
+                # accumulated_reasoning += reasoning_content
+                # 如果设置了 remove_think，跳过 reasoning_content
+                if remove_think:
+                    chunk_idx += 1
+                    continue
 
-            if chunk.choices[0].delta.tool_calls is not None:
-                for tool_call in chunk.choices[0].delta.tool_calls:
-                    if tool_call.function.arguments is None:
+                # 第一次出现 reasoning_content，添加 <think> 开始标签
+                if not thinking_started:
+                    thinking_started = True
+                    pending_content += '<think>\n' + reasoning_content
+                else:
+                    # 继续输出 reasoning_content
+                    pending_content += reasoning_content
+            elif thinking_started and not thinking_ended and delta.get('content'):
+                # reasoning_content 结束，normal content 开始，添加 </think> 结束标签
+                thinking_ended = True
+                pending_content += '\n</think>\n' + delta.get('content')
+
+            if delta.get('content') is not None:
+                pending_content += delta.get('content')
+
+            if delta.get('tool_calls') is not None:
+                for tool_call in delta.get('tool_calls'):
+                    if tool_call['id'] != '':
+                        tool_id = tool_call['id']
+                    if tool_call['function']['name'] is not None:
+                        tool_name = tool_call['function']['name']
+                    if tool_call['function']['arguments'] is None:
                         continue
+                    tool_call['id'] = tool_id
+                    tool_call['name'] = tool_name
                     for tc in tool_calls:
-                        if tc.index == tool_call.index:
-                            tc.function.arguments += tool_call.function.arguments
+                        if tc['index'] == tool_call['index']:
+                            tc['function']['arguments'] += tool_call['function']['arguments']
                             break
                     else:
                         tool_calls.append(tool_call)
 
             if chunk.choices[0].finish_reason is not None:
                 break
+        message_delta['content'] = pending_content
+        message_delta['role'] = 'assistant'
 
-        real_tool_calls = []
-
-        for tc in tool_calls:
-            function = chat_completion_message_tool_call.Function(
-                name=tc.function.name, arguments=tc.function.arguments
-            )
-            real_tool_calls.append(
-                chat_completion_message_tool_call.ChatCompletionMessageToolCall(
-                    id=tc.id, function=function, type='function'
-                )
-            )
-
-        return (
-            chat_completion.ChatCompletion(
-                id=chunk.id,
-                object='chat.completion',
-                created=chunk.created,
-                choices=[
-                    chat_completion.Choice(
-                        index=0,
-                        message=chat_completion.ChatCompletionMessage(
-                            role='assistant',
-                            content=pending_content,
-                            tool_calls=real_tool_calls if len(real_tool_calls) > 0 else None,
-                        ),
-                        finish_reason=chunk.choices[0].finish_reason
-                        if hasattr(chunk.choices[0], 'finish_reason') and chunk.choices[0].finish_reason is not None
-                        else 'stop',
-                        logprobs=chunk.choices[0].logprobs,
-                    )
-                ],
-                model=chunk.model,
-                service_tier=chunk.service_tier if hasattr(chunk, 'service_tier') else None,
-                system_fingerprint=chunk.system_fingerprint if hasattr(chunk, 'system_fingerprint') else None,
-                usage=chunk.usage if hasattr(chunk, 'usage') else None,
-            )
-            if chunk
-            else None
-        )
+        message_delta['tool_calls'] = tool_calls if tool_calls else None
+        return [message_delta]
 
     async def _make_msg(
         self,
-        chat_completion: chat_completion.ChatCompletion,
+        chat_completion: list[dict[str, typing.Any]],
     ) -> provider_message.Message:
-        chatcmpl_message = chat_completion.choices[0].message.dict()
+        chatcmpl_message = chat_completion[0]
 
         # 确保 role 字段存在且不为 None
         if 'role' not in chatcmpl_message or chatcmpl_message['role'] is None:
@@ -130,6 +129,7 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
         use_model: requester.RuntimeLLMModel,
         use_funcs: list[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         self.client.api_key = use_model.token_mgr.get_token()
 
@@ -157,13 +157,145 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
         args['messages'] = messages
 
         # 发送请求
-        resp = await self._req(args, extra_body=extra_args)
+        resp = await self._req(query, args, extra_body=extra_args, remove_think=remove_think)
 
         # 处理请求结果
         message = await self._make_msg(resp)
 
         return message
 
+    async def _req_stream(
+        self,
+        args: dict,
+        extra_body: dict = {},
+    ) -> chat_completion.ChatCompletion:
+        async for chunk in await self.client.chat.completions.create(**args, extra_body=extra_body):
+            yield chunk
+
+    async def _closure_stream(
+        self,
+        query: pipeline_query.Query,
+        req_messages: list[dict],
+        use_model: requester.RuntimeLLMModel,
+        use_funcs: list[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.Message | typing.AsyncGenerator[provider_message.MessageChunk, None]:
+        self.client.api_key = use_model.token_mgr.get_token()
+
+        args = {}
+        args['model'] = use_model.model_entity.name
+
+        if use_funcs:
+            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
+
+            if tools:
+                args['tools'] = tools
+
+        # 设置此次请求中的messages
+        messages = req_messages.copy()
+
+        # 检查vision
+        for msg in messages:
+            if 'content' in msg and isinstance(msg['content'], list):
+                for me in msg['content']:
+                    if me['type'] == 'image_base64':
+                        me['image_url'] = {'url': me['image_base64']}
+                        me['type'] = 'image_url'
+                        del me['image_base64']
+
+        args['messages'] = messages
+        args['stream'] = True
+
+        # 流式处理状态
+        # tool_calls_map: dict[str, provider_message.ToolCall] = {}
+        chunk_idx = 0
+        thinking_started = False
+        thinking_ended = False
+        role = 'assistant'  # 默认角色
+        # accumulated_reasoning = ''  # 仅用于判断何时结束思维链
+
+        async for chunk in self._req_stream(args, extra_body=extra_args):
+            # 解析 chunk 数据
+            if hasattr(chunk, 'choices') and chunk.choices:
+                choice = chunk.choices[0]
+                delta = choice.delta.model_dump() if hasattr(choice, 'delta') else {}
+                finish_reason = getattr(choice, 'finish_reason', None)
+            else:
+                delta = {}
+                finish_reason = None
+
+            # 从第一个 chunk 获取 role，后续使用这个 role
+            if 'role' in delta and delta['role']:
+                role = delta['role']
+
+            # 获取增量内容
+            delta_content = delta.get('content', '')
+            reasoning_content = delta.get('reasoning_content', '')
+
+            # 处理 reasoning_content
+            if reasoning_content:
+                # accumulated_reasoning += reasoning_content
+                # 如果设置了 remove_think，跳过 reasoning_content
+                if remove_think:
+                    chunk_idx += 1
+                    continue
+
+                # 第一次出现 reasoning_content，添加 <think> 开始标签
+                if not thinking_started:
+                    thinking_started = True
+                    delta_content = '<think>\n' + reasoning_content
+                else:
+                    # 继续输出 reasoning_content
+                    delta_content = reasoning_content
+            elif thinking_started and not thinking_ended and delta_content:
+                # reasoning_content 结束，normal content 开始，添加 </think> 结束标签
+                thinking_ended = True
+                delta_content = '\n</think>\n' + delta_content
+
+            # 处理 content 中已有的 <think> 标签（如果需要移除）
+            # if delta_content and remove_think and '<think>' in delta_content:
+            #     import re
+            #
+            #     # 移除 <think> 标签及其内容
+            #     delta_content = re.sub(r'<think>.*?</think>', '', delta_content, flags=re.DOTALL)
+
+            # 处理工具调用增量
+            if delta.get('tool_calls'):
+                for tool_call in delta['tool_calls']:
+                    if tool_call['id'] != '':
+                        tool_id = tool_call['id']
+                    if tool_call['function']['name'] is not None:
+                        tool_name = tool_call['function']['name']
+
+                    if tool_call['type'] is None:
+                        tool_call['type'] = 'function'
+                    tool_call['id'] = tool_id
+                    tool_call['function']['name'] = tool_name
+                    tool_call['function']['arguments'] = (
+                        '' if tool_call['function']['arguments'] is None else tool_call['function']['arguments']
+                    )
+
+            # 跳过空的第一个 chunk（只有 role 没有内容）
+            if chunk_idx == 0 and not delta_content and not reasoning_content and not delta.get('tool_calls'):
+                chunk_idx += 1
+                continue
+
+            # 构建 MessageChunk - 只包含增量内容
+            chunk_data = {
+                'role': role,
+                'content': delta_content if delta_content else None,
+                'tool_calls': delta.get('tool_calls'),
+                'is_final': bool(finish_reason),
+            }
+
+            # 移除 None 值
+            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
+            # return
+
     async def invoke_llm(
         self,
         query: pipeline_query.Query,
@@ -171,6 +303,7 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
         messages: typing.List[provider_message.Message],
         funcs: typing.List[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         req_messages = []  # req_messages 仅用于类内，外部同步由 query.messages 进行
         for m in messages:
@@ -185,7 +318,12 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
 
         try:
             return await self._closure(
-                query=query, req_messages=req_messages, use_model=model, use_funcs=funcs, extra_args=extra_args
+                query=query,
+                req_messages=req_messages,
+                use_model=model,
+                use_funcs=funcs,
+                extra_args=extra_args,
+                remove_think=remove_think,
             )
         except asyncio.TimeoutError:
             raise errors.RequesterError('请求超时')
@@ -202,3 +340,50 @@ class ModelScopeChatCompletions(requester.LLMAPIRequester):
             raise errors.RequesterError(f'请求过于频繁或余额不足: {e.message}')
         except openai.APIError as e:
             raise errors.RequesterError(f'请求错误: {e.message}')
+
+    async def invoke_llm_stream(
+        self,
+        query: pipeline_query.Query,
+        model: requester.RuntimeLLMModel,
+        messages: typing.List[provider_message.Message],
+        funcs: typing.List[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.MessageChunk:
+        req_messages = []  # req_messages 仅用于类内，外部同步由 query.messages 进行
+        for m in messages:
+            msg_dict = m.dict(exclude_none=True)
+            content = msg_dict.get('content')
+            if isinstance(content, list):
+                # 检查 content 列表中是否每个部分都是文本
+                if all(isinstance(part, dict) and part.get('type') == 'text' for part in content):
+                    # 将所有文本部分合并为一个字符串
+                    msg_dict['content'] = '\n'.join(part['text'] for part in content)
+            req_messages.append(msg_dict)
+
+        try:
+            async for item in self._closure_stream(
+                query=query,
+                req_messages=req_messages,
+                use_model=model,
+                use_funcs=funcs,
+                extra_args=extra_args,
+                remove_think=remove_think,
+            ):
+                yield item
+
+        except asyncio.TimeoutError:
+            raise errors.RequesterError('请求超时')
+        except openai.BadRequestError as e:
+            if 'context_length_exceeded' in e.message:
+                raise errors.RequesterError(f'上文过长，请重置会话: {e.message}')
+            else:
+                raise errors.RequesterError(f'请求参数错误: {e.message}')
+        except openai.AuthenticationError as e:
+            raise errors.RequesterError(f'无效的 api-key: {e.message}')
+        except openai.NotFoundError as e:
+            raise errors.RequesterError(f'请求路径错误: {e.message}')
+        except openai.RateLimitError as e:
+            raise errors.RequesterError(f'请求过于频繁或余额不足: {e.message}')
+        except openai.APIError as e:
+            raise errors.RequesterError(f'请求错误: {e.message}')

pkg/provider/modelmgr/requesters/modelscopechatcmpl.yaml

@@ -29,6 +29,8 @@ spec:
       type: int
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./modelscopechatcmpl.py

pkg/provider/modelmgr/requesters/moonshotchatcmpl.py

@@ -25,6 +25,7 @@ class MoonshotChatCompletions(chatcmpl.OpenAIChatCompletions):
         use_model: requester.RuntimeLLMModel,
         use_funcs: list[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         self.client.api_key = use_model.token_mgr.get_token()
 
@@ -54,6 +55,6 @@ class MoonshotChatCompletions(chatcmpl.OpenAIChatCompletions):
         resp = await self._req(args, extra_body=extra_args)
 
         # 处理请求结果
-        message = await self._make_msg(resp)
+        message = await self._make_msg(resp, remove_think)
 
         return message

pkg/provider/modelmgr/requesters/moonshotchatcmpl.yaml

@@ -14,7 +14,7 @@ spec:
         zh_Hans: 基础 URL
       type: string
       required: true
-      default: "https://api.moonshot.com/v1"
+      default: "https://api.moonshot.ai/v1"
     - name: timeout
       label:
         en_US: Timeout
@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./moonshotchatcmpl.py

pkg/provider/modelmgr/requesters/newapichatcmpl.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+import typing
+import openai
+
+from . import chatcmpl
+
+
+class NewAPIChatCompletions(chatcmpl.OpenAIChatCompletions):
+    """New API ChatCompletion API 请求器"""
+
+    client: openai.AsyncClient
+
+    default_config: dict[str, typing.Any] = {
+        'base_url': 'http://localhost:3000/v1',
+        'timeout': 120,
+    }

pkg/provider/modelmgr/requesters/newapichatcmpl.yaml

@@ -0,0 +1,31 @@
+apiVersion: v1
+kind: LLMAPIRequester
+metadata:
+  name: new-api-chat-completions
+  label:
+    en_US: New API
+    zh_Hans: New API
+  icon: newapi.png
+spec:
+  config:
+    - name: base_url
+      label:
+        en_US: Base URL
+        zh_Hans: 基础 URL
+      type: string
+      required: true
+      default: "http://localhost:3000/v1"
+    - name: timeout
+      label:
+        en_US: Timeout
+        zh_Hans: 超时时间
+      type: integer
+      required: true
+      default: 120
+  support_type:
+    - llm
+    - text-embedding
+execution:
+  python:
+    path: ./newapichatcmpl.py
+    attr: NewAPIChatCompletions

pkg/provider/modelmgr/requesters/ollamachat.py

@@ -17,7 +17,7 @@ import langbot_plugin.api.entities.builtin.provider.message as provider_message
 REQUESTER_NAME: str = 'ollama-chat'
 
 
-class OllamaChatCompletions(requester.LLMAPIRequester):
+class OllamaChatCompletions(requester.ProviderAPIRequester):
     """Ollama平台 ChatCompletion API请求器"""
 
     client: ollama.AsyncClient
@@ -44,6 +44,7 @@ class OllamaChatCompletions(requester.LLMAPIRequester):
         use_model: requester.RuntimeLLMModel,
         use_funcs: list[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         args = extra_args.copy()
         args['model'] = use_model.model_entity.name
@@ -110,6 +111,7 @@ class OllamaChatCompletions(requester.LLMAPIRequester):
         messages: typing.List[provider_message.Message],
         funcs: typing.List[resource_tool.LLMTool] = None,
         extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
     ) -> provider_message.Message:
         req_messages: list = []
         for m in messages:
@@ -126,6 +128,21 @@ class OllamaChatCompletions(requester.LLMAPIRequester):
                 use_model=model,
                 use_funcs=funcs,
                 extra_args=extra_args,
+                remove_think=remove_think,
             )
         except asyncio.TimeoutError:
             raise errors.RequesterError('请求超时')
+
+    async def invoke_embedding(
+        self,
+        model: requester.RuntimeEmbeddingModel,
+        input_text: list[str],
+        extra_args: dict[str, typing.Any] = {},
+    ) -> list[list[float]]:
+        return (
+            await self.client.embed(
+                model=model.model_entity.name,
+                input=input_text,
+                **extra_args,
+            )
+        ).embeddings

pkg/provider/modelmgr/requesters/ollamachat.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./ollamachat.py

pkg/provider/modelmgr/requesters/openrouterchatcmpl.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./openrouterchatcmpl.py

pkg/provider/modelmgr/requesters/ppiochatcmpl.py

@@ -4,6 +4,12 @@ import openai
 import typing
 
 from . import chatcmpl
+from .. import requester
+import openai.types.chat.chat_completion as chat_completion
+import re
+import langbot_plugin.api.entities.builtin.provider.message as provider_message
+import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
+import langbot_plugin.api.entities.builtin.resource.tool as resource_tool
 
 
 class PPIOChatCompletions(chatcmpl.OpenAIChatCompletions):
@@ -15,3 +21,188 @@ class PPIOChatCompletions(chatcmpl.OpenAIChatCompletions):
         'base_url': 'https://api.ppinfra.com/v3/openai',
         'timeout': 120,
     }
+
+    is_think: bool = False
+
+    async def _make_msg(
+        self,
+        chat_completion: chat_completion.ChatCompletion,
+        remove_think: bool,
+    ) -> provider_message.Message:
+        chatcmpl_message = chat_completion.choices[0].message.model_dump()
+        # print(chatcmpl_message.keys(), chatcmpl_message.values())
+
+        # 确保 role 字段存在且不为 None
+        if 'role' not in chatcmpl_message or chatcmpl_message['role'] is None:
+            chatcmpl_message['role'] = 'assistant'
+
+        reasoning_content = chatcmpl_message['reasoning_content'] if 'reasoning_content' in chatcmpl_message else None
+
+        # deepseek的reasoner模型
+        chatcmpl_message['content'] = await self._process_thinking_content(
+            chatcmpl_message['content'], reasoning_content, remove_think
+        )
+
+        # 移除 reasoning_content 字段，避免传递给 Message
+        if 'reasoning_content' in chatcmpl_message:
+            del chatcmpl_message['reasoning_content']
+
+        message = provider_message.Message(**chatcmpl_message)
+
+        return message
+
+    async def _process_thinking_content(
+        self,
+        content: str,
+        reasoning_content: str = None,
+        remove_think: bool = False,
+    ) -> tuple[str, str]:
+        """处理思维链内容
+
+        Args:
+            content: 原始内容
+            reasoning_content: reasoning_content 字段内容
+            remove_think: 是否移除思维链
+
+        Returns:
+            处理后的内容
+        """
+        if remove_think:
+            content = re.sub(r'<think>.*?</think>', '', content, flags=re.DOTALL)
+        else:
+            if reasoning_content is not None:
+                content = '<think>\n' + reasoning_content + '\n</think>\n' + content
+        return content
+
+    async def _make_msg_chunk(
+        self,
+        delta: dict[str, typing.Any],
+        idx: int,
+    ) -> provider_message.MessageChunk:
+        # 处理流式chunk和完整响应的差异
+        # print(chat_completion.choices[0])
+
+        # 确保 role 字段存在且不为 None
+        if 'role' not in delta or delta['role'] is None:
+            delta['role'] = 'assistant'
+
+        reasoning_content = delta['reasoning_content'] if 'reasoning_content' in delta else None
+
+        delta['content'] = '' if delta['content'] is None else delta['content']
+        # print(reasoning_content)
+
+        # deepseek的reasoner模型
+
+        if reasoning_content is not None:
+            delta['content'] += reasoning_content
+
+        message = provider_message.MessageChunk(**delta)
+
+        return message
+
+    async def _closure_stream(
+        self,
+        query: pipeline_query.Query,
+        req_messages: list[dict],
+        use_model: requester.RuntimeLLMModel,
+        use_funcs: list[resource_tool.LLMTool] = None,
+        extra_args: dict[str, typing.Any] = {},
+        remove_think: bool = False,
+    ) -> provider_message.Message | typing.AsyncGenerator[provider_message.MessageChunk, None]:
+        self.client.api_key = use_model.token_mgr.get_token()
+
+        args = {}
+        args['model'] = use_model.model_entity.name
+
+        if use_funcs:
+            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)
+
+            if tools:
+                args['tools'] = tools
+
+        # 设置此次请求中的messages
+        messages = req_messages.copy()
+
+        # 检查vision
+        for msg in messages:
+            if 'content' in msg and isinstance(msg['content'], list):
+                for me in msg['content']:
+                    if me['type'] == 'image_base64':
+                        me['image_url'] = {'url': me['image_base64']}
+                        me['type'] = 'image_url'
+                        del me['image_base64']
+
+        args['messages'] = messages
+        args['stream'] = True
+
+        # tool_calls_map: dict[str, provider_message.ToolCall] = {}
+        chunk_idx = 0
+        thinking_started = False
+        thinking_ended = False
+        role = 'assistant'  # 默认角色
+        async for chunk in self._req_stream(args, extra_body=extra_args):
+            # 解析 chunk 数据
+            if hasattr(chunk, 'choices') and chunk.choices:
+                choice = chunk.choices[0]
+                delta = choice.delta.model_dump() if hasattr(choice, 'delta') else {}
+                finish_reason = getattr(choice, 'finish_reason', None)
+            else:
+                delta = {}
+                finish_reason = None
+
+            # 从第一个 chunk 获取 role，后续使用这个 role
+            if 'role' in delta and delta['role']:
+                role = delta['role']
+
+            # 获取增量内容
+            delta_content = delta.get('content', '')
+            # reasoning_content = delta.get('reasoning_content', '')
+
+            if remove_think:
+                if delta['content'] is not None:
+                    if '<think>' in delta['content'] and not thinking_started and not thinking_ended:
+                        thinking_started = True
+                        continue
+                    elif delta['content'] == r'</think>' and not thinking_ended:
+                        thinking_ended = True
+                        continue
+                    elif thinking_ended and delta['content'] == '\n\n' and thinking_started:
+                        thinking_started = False
+                        continue
+                    elif thinking_started and not thinking_ended:
+                        continue
+
+            # delta_tool_calls = None
+            if delta.get('tool_calls'):
+                for tool_call in delta['tool_calls']:
+                    if tool_call['id'] and tool_call['function']['name']:
+                        tool_id = tool_call['id']
+                        tool_name = tool_call['function']['name']
+
+                    if tool_call['id'] is None:
+                        tool_call['id'] = tool_id
+                    if tool_call['function']['name'] is None:
+                        tool_call['function']['name'] = tool_name
+                    if tool_call['function']['arguments'] is None:
+                        tool_call['function']['arguments'] = ''
+                    if tool_call['type'] is None:
+                        tool_call['type'] = 'function'
+
+            # 跳过空的第一个 chunk（只有 role 没有内容）
+            if chunk_idx == 0 and not delta_content and not delta.get('tool_calls'):
+                chunk_idx += 1
+                continue
+
+            # 构建 MessageChunk - 只包含增量内容
+            chunk_data = {
+                'role': role,
+                'content': delta_content if delta_content else None,
+                'tool_calls': delta.get('tool_calls'),
+                'is_final': bool(finish_reason),
+            }
+
+            # 移除 None 值
+            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

pkg/provider/modelmgr/requesters/ppiochatcmpl.yaml

@@ -29,6 +29,9 @@ spec:
       type: int
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./ppiochatcmpl.py

pkg/provider/modelmgr/requesters/qhaigcchatcmpl.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+import openai
+import typing
+
+from . import chatcmpl
+
+
+class QHAIGCChatCompletions(chatcmpl.OpenAIChatCompletions):
+    """启航 AI ChatCompletion API 请求器"""
+
+    client: openai.AsyncClient
+
+    default_config: dict[str, typing.Any] = {
+        'base_url': 'https://api.qhaigc.com/v1',
+        'timeout': 120,
+    }

pkg/provider/modelmgr/requesters/qhaigcchatcmpl.yaml

@@ -0,0 +1,38 @@
+apiVersion: v1
+kind: LLMAPIRequester
+metadata:
+  name: qhaigc-chat-completions
+  label:
+    en_US: QH AI
+    zh_Hans: 启航 AI
+  icon: qhaigc.png
+spec:
+  config:
+    - name: base_url
+      label:
+        en_US: Base URL
+        zh_Hans: 基础 URL
+      type: string
+      required: true
+      default: "https://api.qhaigc.net/v1"
+    - name: args
+      label:
+        en_US: Args
+        zh_Hans: 附加参数
+      type: object
+      required: true
+      default: {}
+    - name: timeout
+      label:
+        en_US: Timeout
+        zh_Hans: 超时时间
+      type: int
+      required: true
+      default: 120
+  support_type:
+    - llm
+    - text-embedding
+execution:
+  python:
+    path: ./qhaigcchatcmpl.py
+    attr: QHAIGCChatCompletions

pkg/provider/modelmgr/requesters/shengsuanyun.py

@@ -0,0 +1,32 @@
+from __future__ import annotations
+
+import openai
+import typing
+
+from . import chatcmpl
+import openai.types.chat.chat_completion as chat_completion
+
+
+class ShengSuanYunChatCompletions(chatcmpl.OpenAIChatCompletions):
+    """胜算云(ModelSpot.AI) ChatCompletion API 请求器"""
+
+    client: openai.AsyncClient
+
+    default_config: dict[str, typing.Any] = {
+        'base_url': 'https://router.shengsuanyun.com/api/v1',
+        'timeout': 120,
+    }
+
+    async def _req(
+        self,
+        args: dict,
+        extra_body: dict = {},
+    ) -> chat_completion.ChatCompletion:
+        return await self.client.chat.completions.create(
+            **args,
+            extra_body=extra_body,
+            extra_headers={
+                'HTTP-Referer': 'https://langbot.app',
+                'X-Title': 'LangBot',
+            },
+        )

pkg/provider/modelmgr/requesters/shengsuanyun.yaml

@@ -0,0 +1,38 @@
+apiVersion: v1
+kind: LLMAPIRequester
+metadata:
+  name: shengsuanyun-chat-completions
+  label:
+    en_US: ShengSuanYun
+    zh_Hans: 胜算云
+  icon: shengsuanyun.svg
+spec:
+  config:
+    - name: base_url
+      label:
+        en_US: Base URL
+        zh_Hans: 基础 URL
+      type: string
+      required: true
+      default: "https://router.shengsuanyun.com/api/v1"
+    - name: args
+      label:
+        en_US: Args
+        zh_Hans: 附加参数
+      type: object
+      required: true
+      default: {}
+    - name: timeout
+      label:
+        en_US: Timeout
+        zh_Hans: 超时时间
+      type: int
+      required: true
+      default: 120
+  support_type:
+    - llm
+    - text-embedding
+execution:
+  python:
+    path: ./shengsuanyun.py
+    attr: ShengSuanYunChatCompletions

pkg/provider/modelmgr/requesters/siliconflowchatcmpl.yaml

@@ -22,6 +22,9 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
+    - text-embedding
 execution:
   python:
     path: ./siliconflowchatcmpl.py

pkg/provider/modelmgr/requesters/volcarkchatcmpl.yaml

@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./volcarkchatcmpl.py

pkg/provider/modelmgr/requesters/xaichatcmpl.yaml

@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./xaichatcmpl.py

pkg/provider/modelmgr/requesters/zhipuaichatcmpl.yaml

@@ -22,6 +22,8 @@ spec:
       type: integer
       required: true
       default: 120
+  support_type:
+    - llm
 execution:
   python:
     path: ./zhipuaichatcmpl.py

pkg/provider/runners/dashscopeapi.py

@@ -102,8 +102,14 @@ class DashScopeAPIRunner(runner.RequestRunner):
         plain_text = ''  # 用户输入的纯文本信息
         image_ids = []  # 用户输入的图片ID列表 （暂不支持）
 
-        plain_text, image_ids = await self._preprocess_user_message(query)
+        think_start = False
+        think_end = False
 
+        plain_text, image_ids = await self._preprocess_user_message(query)
+        has_thoughts = True  # 获取思考过程
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+        if remove_think:
+            has_thoughts = False
         # 发送对话请求
         response = dashscope.Application.call(
             api_key=self.api_key,  # 智能体应用的API Key
@@ -112,43 +118,108 @@ class DashScopeAPIRunner(runner.RequestRunner):
             stream=True,  # 流式输出
             incremental_output=True,  # 增量输出，使用流式输出需要开启增量输出
             session_id=query.session.using_conversation.uuid,  # 会话ID用于，多轮对话
+            has_thoughts=has_thoughts,
             # rag_options={                                     # 主要用于文件交互，暂不支持
             #     "session_file_ids": ["FILE_ID1"],             # FILE_ID1 替换为实际的临时文件ID,逗号隔开多个
             # }
         )
+        idx_chunk = 0
+        try:
+            is_stream = await query.adapter.is_stream_output_supported()
 
-        for chunk in response:
-            if chunk.get('status_code') != 200:
-                raise DashscopeAPIError(
-                    f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
-                )
-            if not chunk:
-                continue
+        except AttributeError:
+            is_stream = False
+        if is_stream:
+            for chunk in response:
+                if chunk.get('status_code') != 200:
+                    raise DashscopeAPIError(
+                        f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
+                    )
+                if not chunk:
+                    continue
+                idx_chunk += 1
+                # 获取流式传输的output
+                stream_output = chunk.get('output', {})
+                stream_think = stream_output.get('thoughts', [])
+                if stream_think[0].get('thought'):
+                    if not think_start:
+                        think_start = True
+                        pending_content += f'<think>\n{stream_think[0].get("thought")}'
+                    else:
+                        # 继续输出 reasoning_content
+                        pending_content += stream_think[0].get('thought')
+                elif stream_think[0].get('thought') == '' and not think_end:
+                    think_end = True
+                    pending_content += '\n</think>\n'
+                if stream_output.get('text') is not None:
+                    pending_content += stream_output.get('text')
+                # 是否是流式最后一个chunk
+                is_final = False if stream_output.get('finish_reason', False) == 'null' else True
 
-            # 获取流式传输的output
-            stream_output = chunk.get('output', {})
-            if stream_output.get('text') is not None:
-                pending_content += stream_output.get('text')
+                # 获取模型传出的参考资料列表
+                references_dict_list = stream_output.get('doc_references', [])
 
-        # 保存当前会话的session_id用于下次对话的语境
-        query.session.using_conversation.uuid = stream_output.get('session_id')
+                # 从模型传出的参考资料信息中提取用于替换的字典
+                if references_dict_list is not None:
+                    for doc in references_dict_list:
+                        if doc.get('index_id') is not None:
+                            references_dict[doc.get('index_id')] = doc.get('doc_name')
 
-        # 获取模型传出的参考资料列表
-        references_dict_list = stream_output.get('doc_references', [])
+                    # 将参考资料替换到文本中
+                    pending_content = self._replace_references(pending_content, references_dict)
 
-        # 从模型传出的参考资料信息中提取用于替换的字典
-        if references_dict_list is not None:
-            for doc in references_dict_list:
-                if doc.get('index_id') is not None:
-                    references_dict[doc.get('index_id')] = doc.get('doc_name')
+                if idx_chunk % 8 == 0 or is_final:
+                    yield provider_message.MessageChunk(
+                        role='assistant',
+                        content=pending_content,
+                        is_final=is_final,
+                    )
+            # 保存当前会话的session_id用于下次对话的语境
+            query.session.using_conversation.uuid = stream_output.get('session_id')
+        else:
+            for chunk in response:
+                if chunk.get('status_code') != 200:
+                    raise DashscopeAPIError(
+                        f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
+                    )
+                if not chunk:
+                    continue
+                idx_chunk += 1
+                # 获取流式传输的output
+                stream_output = chunk.get('output', {})
+                stream_think = stream_output.get('thoughts', [])
+                if stream_think[0].get('thought'):
+                    if not think_start:
+                        think_start = True
+                        pending_content += f'<think>\n{stream_think[0].get("thought")}'
+                    else:
+                        # 继续输出 reasoning_content
+                        pending_content += stream_think[0].get('thought')
+                elif stream_think[0].get('thought') == '' and not think_end:
+                    think_end = True
+                    pending_content += '\n</think>\n'
+                if stream_output.get('text') is not None:
+                    pending_content += stream_output.get('text')
 
-            # 将参考资料替换到文本中
-            pending_content = self._replace_references(pending_content, references_dict)
+            # 保存当前会话的session_id用于下次对话的语境
+            query.session.using_conversation.uuid = stream_output.get('session_id')
 
-        yield provider_message.Message(
-            role='assistant',
-            content=pending_content,
-        )
+            # 获取模型传出的参考资料列表
+            references_dict_list = stream_output.get('doc_references', [])
+
+            # 从模型传出的参考资料信息中提取用于替换的字典
+            if references_dict_list is not None:
+                for doc in references_dict_list:
+                    if doc.get('index_id') is not None:
+                        references_dict[doc.get('index_id')] = doc.get('doc_name')
+
+                # 将参考资料替换到文本中
+                pending_content = self._replace_references(pending_content, references_dict)
+
+            yield provider_message.Message(
+                role='assistant',
+                content=pending_content,
+            )
 
     async def _workflow_messages(
         self, query: pipeline_query.Query
@@ -176,52 +247,108 @@ class DashScopeAPIRunner(runner.RequestRunner):
             incremental_output=True,  # 增量输出，使用流式输出需要开启增量输出
             session_id=query.session.using_conversation.uuid,  # 会话ID用于，多轮对话
             biz_params=biz_params,  # 工作流应用的自定义输入参数传递
+            flow_stream_mode='message_format',  # 消息模式，输出/结束节点的流式结果
             # rag_options={                                     # 主要用于文件交互，暂不支持
             #     "session_file_ids": ["FILE_ID1"],             # FILE_ID1 替换为实际的临时文件ID,逗号隔开多个
             # }
         )
 
         # 处理API返回的流式输出
-        for chunk in response:
-            if chunk.get('status_code') != 200:
-                raise DashscopeAPIError(
-                    f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
-                )
-            if not chunk:
-                continue
+        try:
+            is_stream = await query.adapter.is_stream_output_supported()
 
-            # 获取流式传输的output
-            stream_output = chunk.get('output', {})
-            if stream_output.get('text') is not None:
-                pending_content += stream_output.get('text')
+        except AttributeError:
+            is_stream = False
+        idx_chunk = 0
+        if is_stream:
+            for chunk in response:
+                if chunk.get('status_code') != 200:
+                    raise DashscopeAPIError(
+                        f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
+                    )
+                if not chunk:
+                    continue
+                idx_chunk += 1
+                # 获取流式传输的output
+                stream_output = chunk.get('output', {})
+                if stream_output.get('workflow_message') is not None:
+                    pending_content += stream_output.get('workflow_message').get('message').get('content')
+                # if stream_output.get('text') is not None:
+                #     pending_content += stream_output.get('text')
 
-        # 保存当前会话的session_id用于下次对话的语境
-        query.session.using_conversation.uuid = stream_output.get('session_id')
+                is_final = False if stream_output.get('finish_reason', False) == 'null' else True
 
-        # 获取模型传出的参考资料列表
-        references_dict_list = stream_output.get('doc_references', [])
+                # 获取模型传出的参考资料列表
+                references_dict_list = stream_output.get('doc_references', [])
 
-        # 从模型传出的参考资料信息中提取用于替换的字典
-        if references_dict_list is not None:
-            for doc in references_dict_list:
-                if doc.get('index_id') is not None:
-                    references_dict[doc.get('index_id')] = doc.get('doc_name')
+                # 从模型传出的参考资料信息中提取用于替换的字典
+                if references_dict_list is not None:
+                    for doc in references_dict_list:
+                        if doc.get('index_id') is not None:
+                            references_dict[doc.get('index_id')] = doc.get('doc_name')
 
-            # 将参考资料替换到文本中
-            pending_content = self._replace_references(pending_content, references_dict)
+                    # 将参考资料替换到文本中
+                    pending_content = self._replace_references(pending_content, references_dict)
+                if idx_chunk % 8 == 0 or is_final:
+                    yield provider_message.MessageChunk(
+                        role='assistant',
+                        content=pending_content,
+                        is_final=is_final,
+                    )
 
-        yield provider_message.Message(
-            role='assistant',
-            content=pending_content,
-        )
+            # 保存当前会话的session_id用于下次对话的语境
+            query.session.using_conversation.uuid = stream_output.get('session_id')
+
+        else:
+            for chunk in response:
+                if chunk.get('status_code') != 200:
+                    raise DashscopeAPIError(
+                        f'Dashscope API 请求失败: status_code={chunk.get("status_code")} message={chunk.get("message")} request_id={chunk.get("request_id")} '
+                    )
+                if not chunk:
+                    continue
+
+                # 获取流式传输的output
+                stream_output = chunk.get('output', {})
+                if stream_output.get('text') is not None:
+                    pending_content += stream_output.get('text')
+
+                is_final = False if stream_output.get('finish_reason', False) == 'null' else True
+
+            # 保存当前会话的session_id用于下次对话的语境
+            query.session.using_conversation.uuid = stream_output.get('session_id')
+
+            # 获取模型传出的参考资料列表
+            references_dict_list = stream_output.get('doc_references', [])
+
+            # 从模型传出的参考资料信息中提取用于替换的字典
+            if references_dict_list is not None:
+                for doc in references_dict_list:
+                    if doc.get('index_id') is not None:
+                        references_dict[doc.get('index_id')] = doc.get('doc_name')
+
+                # 将参考资料替换到文本中
+                pending_content = self._replace_references(pending_content, references_dict)
+
+            yield provider_message.Message(
+                role='assistant',
+                content=pending_content,
+            )
 
     async def run(self, query: pipeline_query.Query) -> typing.AsyncGenerator[provider_message.Message, None]:
         """运行"""
+        msg_seq = 0
         if self.app_type == 'agent':
             async for msg in self._agent_messages(query):
+                if isinstance(msg, provider_message.MessageChunk):
+                    msg_seq += 1
+                    msg.msg_sequence = msg_seq
                 yield msg
         elif self.app_type == 'workflow':
             async for msg in self._workflow_messages(query):
+                if isinstance(msg, provider_message.MessageChunk):
+                    msg_seq += 1
+                    msg.msg_sequence = msg_seq
                 yield msg
         else:
             raise DashscopeAPIError(f'不支持的 Dashscope 应用类型: {self.app_type}')

pkg/provider/runners/difysvapi.py

@@ -3,7 +3,6 @@ from __future__ import annotations
 import typing
 import json
 import uuid
-import re
 import base64
 
 
@@ -38,29 +37,38 @@ class DifyServiceAPIRunner(runner.RequestRunner):
             base_url=self.pipeline_config['ai']['dify-service-api']['base-url'],
         )
 
-    def _try_convert_thinking(self, resp_text: str) -> str:
-        """尝试转换 Dify 的思考提示"""
-        if not resp_text.startswith(
-            '<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>'
-        ):
-            return resp_text
+    def _process_thinking_content(
+        self,
+        content: str,
+    ) -> tuple[str, str]:
+        """处理思维链内容
 
-        if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'original':
-            return resp_text
+        Args:
+            content: 原始内容
+        Returns:
+            (处理后的内容, 提取的思维链内容)
+        """
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+        thinking_content = ''
+        # 从 content 中提取 <think> 标签内容
+        if content and '<think>' in content and '</think>' in content:
+            import re
 
-        if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'remove':
-            return re.sub(
-                r'<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>.*?</details>',
-                '',
-                resp_text,
-                flags=re.DOTALL,
-            )
+            think_pattern = r'<think>(.*?)</think>'
+            think_matches = re.findall(think_pattern, content, re.DOTALL)
+            if think_matches:
+                thinking_content = '\n'.join(think_matches)
+                # 移除 content 中的 <think> 标签
+                content = re.sub(think_pattern, '', content, flags=re.DOTALL).strip()
 
-        if self.pipeline_config['ai']['dify-service-api']['thinking-convert'] == 'plain':
-            pattern = r'<details style="color:gray;background-color: #f8f8f8;padding: 8px;border-radius: 4px;" open> <summary> Thinking... </summary>(.*?)</details>'
-            thinking_text = re.search(pattern, resp_text, flags=re.DOTALL)
-            content_text = re.sub(pattern, '', resp_text, flags=re.DOTALL)
-            return f'<think>{thinking_text.group(1)}</think>\n{content_text}'
+        # 3. 根据 remove_think 参数决定是否保留思维链
+        if remove_think:
+            return content, ''
+        else:
+            # 如果有思维链内容，将其以 <think> 格式添加到 content 开头
+            if thinking_content:
+                content = f'<think>\n{thinking_content}\n</think>\n{content}'.strip()
+            return content, thinking_content
 
     async def _preprocess_user_message(self, query: pipeline_query.Query) -> tuple[str, list[str]]:
         """预处理用户消息，提取纯文本，并将图片上传到 Dify 服务
@@ -134,17 +142,20 @@ class DifyServiceAPIRunner(runner.RequestRunner):
             if mode == 'workflow':
                 if chunk['event'] == 'node_finished':
                     if chunk['data']['node_type'] == 'answer':
+                        content, _ = self._process_thinking_content(chunk['data']['outputs']['answer'])
+
                         yield provider_message.Message(
                             role='assistant',
-                            content=self._try_convert_thinking(chunk['data']['outputs']['answer']),
+                            content=content,
                         )
             elif mode == 'basic':
                 if chunk['event'] == 'message':
                     basic_mode_pending_chunk += chunk['answer']
                 elif chunk['event'] == 'message_end':
+                    content, _ = self._process_thinking_content(basic_mode_pending_chunk)
                     yield provider_message.Message(
                         role='assistant',
-                        content=self._try_convert_thinking(basic_mode_pending_chunk),
+                        content=content,
                     )
                     basic_mode_pending_chunk = ''
 
@@ -195,14 +206,15 @@ class DifyServiceAPIRunner(runner.RequestRunner):
             if chunk['event'] in ignored_events:
                 continue
 
-            if chunk['event'] == 'agent_message':
+            if chunk['event'] == 'agent_message' or chunk['event'] == 'message':
                 pending_agent_message += chunk['answer']
             else:
                 if pending_agent_message.strip() != '':
                     pending_agent_message = pending_agent_message.replace('</details>Action:', '</details>')
+                    content, _ = self._process_thinking_content(pending_agent_message)
                     yield provider_message.Message(
                         role='assistant',
-                        content=self._try_convert_thinking(pending_agent_message),
+                        content=content,
                     )
                 pending_agent_message = ''
 
@@ -312,26 +324,352 @@ class DifyServiceAPIRunner(runner.RequestRunner):
             elif chunk['event'] == 'workflow_finished':
                 if chunk['data']['error']:
                     raise errors.DifyAPIError(chunk['data']['error'])
+                content, _ = self._process_thinking_content(chunk['data']['outputs']['summary'])
 
                 msg = provider_message.Message(
                     role='assistant',
-                    content=chunk['data']['outputs']['summary'],
+                    content=content,
                 )
 
                 yield msg
 
+    async def _chat_messages_chunk(
+        self, query: pipeline_query.Query
+    ) -> typing.AsyncGenerator[provider_message.MessageChunk, None]:
+        """调用聊天助手"""
+        cov_id = query.session.using_conversation.uuid or ''
+        query.variables['conversation_id'] = cov_id
+
+        plain_text, image_ids = await self._preprocess_user_message(query)
+
+        files = [
+            {
+                'type': 'image',
+                'transfer_method': 'local_file',
+                'upload_file_id': image_id,
+            }
+            for image_id in image_ids
+        ]
+
+        basic_mode_pending_chunk = ''
+
+        inputs = {}
+
+        inputs.update(query.variables)
+        message_idx = 0
+
+        chunk = None  # 初始化chunk变量，防止在没有响应时引用错误
+
+        is_final = False
+        think_start = False
+        think_end = False
+
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+
+        async for chunk in self.dify_client.chat_messages(
+            inputs=inputs,
+            query=plain_text,
+            user=f'{query.session.launcher_type.value}_{query.session.launcher_id}',
+            conversation_id=cov_id,
+            files=files,
+            timeout=120,
+        ):
+            self.ap.logger.debug('dify-chat-chunk: ' + str(chunk))
+
+            # if chunk['event'] == 'workflow_started':
+            #     mode = 'workflow'
+            # if mode == 'workflow':
+            # elif mode == 'basic':
+            # 因为都只是返回的 message也没有工具调用什么的，暂时不分类
+            if chunk['event'] == 'message':
+                message_idx += 1
+                if remove_think:
+                    if '<think>' in chunk['answer'] and not think_start:
+                        think_start = True
+                        continue
+                    if '</think>' in chunk['answer'] and not think_end:
+                        import re
+
+                        content = re.sub(r'^\n</think>', '', chunk['answer'])
+                        basic_mode_pending_chunk += content
+                        think_end = True
+                    elif think_end:
+                        basic_mode_pending_chunk += chunk['answer']
+                    if think_start:
+                        continue
+
+                else:
+                    basic_mode_pending_chunk += chunk['answer']
+
+            if chunk['event'] == 'message_end':
+                is_final = True
+
+            if is_final or message_idx % 8 == 0:
+                # content, _ = self._process_thinking_content(basic_mode_pending_chunk)
+                yield provider_message.MessageChunk(
+                    role='assistant',
+                    content=basic_mode_pending_chunk,
+                    is_final=is_final,
+                )
+
+        if chunk is None:
+            raise errors.DifyAPIError('Dify API 没有返回任何响应，请检查网络连接和API配置')
+
+        query.session.using_conversation.uuid = chunk['conversation_id']
+
+    async def _agent_chat_messages_chunk(
+        self, query: pipeline_query.Query
+    ) -> typing.AsyncGenerator[provider_message.MessageChunk, None]:
+        """调用聊天助手"""
+        cov_id = query.session.using_conversation.uuid or ''
+        query.variables['conversation_id'] = cov_id
+
+        plain_text, image_ids = await self._preprocess_user_message(query)
+
+        files = [
+            {
+                'type': 'image',
+                'transfer_method': 'local_file',
+                'upload_file_id': image_id,
+            }
+            for image_id in image_ids
+        ]
+
+        ignored_events = []
+
+        inputs = {}
+
+        inputs.update(query.variables)
+
+        pending_agent_message = ''
+
+        chunk = None  # 初始化chunk变量，防止在没有响应时引用错误
+        message_idx = 0
+        is_final = False
+        think_start = False
+        think_end = False
+
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+
+        async for chunk in self.dify_client.chat_messages(
+            inputs=inputs,
+            query=plain_text,
+            user=f'{query.session.launcher_type.value}_{query.session.launcher_id}',
+            response_mode='streaming',
+            conversation_id=cov_id,
+            files=files,
+            timeout=120,
+        ):
+            self.ap.logger.debug('dify-agent-chunk: ' + str(chunk))
+
+            if chunk['event'] in ignored_events:
+                continue
+
+            if chunk['event'] == 'agent_message':
+                message_idx += 1
+                if remove_think:
+                    if '<think>' in chunk['answer'] and not think_start:
+                        think_start = True
+                        continue
+                    if '</think>' in chunk['answer'] and not think_end:
+                        import re
+
+                        content = re.sub(r'^\n</think>', '', chunk['answer'])
+                        pending_agent_message += content
+                        think_end = True
+                    elif think_end or not think_start:
+                        pending_agent_message += chunk['answer']
+                    if think_start:
+                        continue
+
+                else:
+                    pending_agent_message += chunk['answer']
+            elif chunk['event'] == 'message_end':
+                is_final = True
+            else:
+                if chunk['event'] == 'agent_thought':
+                    if chunk['tool'] != '' and chunk['observation'] != '':  # 工具调用结果，跳过
+                        continue
+                    message_idx += 1
+                    if chunk['tool']:
+                        msg = provider_message.MessageChunk(
+                            role='assistant',
+                            tool_calls=[
+                                provider_message.ToolCall(
+                                    id=chunk['id'],
+                                    type='function',
+                                    function=provider_message.FunctionCall(
+                                        name=chunk['tool'],
+                                        arguments=json.dumps({}),
+                                    ),
+                                )
+                            ],
+                        )
+                        yield msg
+                if chunk['event'] == 'message_file':
+                    message_idx += 1
+                    if chunk['type'] == 'image' and chunk['belongs_to'] == 'assistant':
+                        base_url = self.dify_client.base_url
+
+                        if base_url.endswith('/v1'):
+                            base_url = base_url[:-3]
+
+                        image_url = base_url + chunk['url']
+
+                        yield provider_message.MessageChunk(
+                            role='assistant',
+                            content=[provider_message.ContentElement.from_image_url(image_url)],
+                            is_final=is_final,
+                        )
+
+                if chunk['event'] == 'error':
+                    raise errors.DifyAPIError('dify 服务错误: ' + chunk['message'])
+            if message_idx % 8 == 0 or is_final:
+                yield provider_message.MessageChunk(
+                    role='assistant',
+                    content=pending_agent_message,
+                    is_final=is_final,
+                )
+
+        if chunk is None:
+            raise errors.DifyAPIError('Dify API 没有返回任何响应，请检查网络连接和API配置')
+
+        query.session.using_conversation.uuid = chunk['conversation_id']
+
+    async def _workflow_messages_chunk(
+        self, query: pipeline_query.Query
+    ) -> typing.AsyncGenerator[provider_message.MessageChunk, None]:
+        """调用工作流"""
+
+        if not query.session.using_conversation.uuid:
+            query.session.using_conversation.uuid = str(uuid.uuid4())
+
+        query.variables['conversation_id'] = query.session.using_conversation.uuid
+
+        plain_text, image_ids = await self._preprocess_user_message(query)
+
+        files = [
+            {
+                'type': 'image',
+                'transfer_method': 'local_file',
+                'upload_file_id': image_id,
+            }
+            for image_id in image_ids
+        ]
+
+        ignored_events = ['workflow_started']
+
+        inputs = {  # these variables are legacy variables, we need to keep them for compatibility
+            'langbot_user_message_text': plain_text,
+            'langbot_session_id': query.variables['session_id'],
+            'langbot_conversation_id': query.variables['conversation_id'],
+            'langbot_msg_create_time': query.variables['msg_create_time'],
+        }
+
+        inputs.update(query.variables)
+        messsage_idx = 0
+        is_final = False
+        think_start = False
+        think_end = False
+        workflow_contents = ''
+
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+        async for chunk in self.dify_client.workflow_run(
+            inputs=inputs,
+            user=f'{query.session.launcher_type.value}_{query.session.launcher_id}',
+            files=files,
+            timeout=120,
+        ):
+            self.ap.logger.debug('dify-workflow-chunk: ' + str(chunk))
+            if chunk['event'] in ignored_events:
+                continue
+            if chunk['event'] == 'workflow_finished':
+                is_final = True
+                if chunk['data']['error']:
+                    raise errors.DifyAPIError(chunk['data']['error'])
+
+            if chunk['event'] == 'text_chunk':
+                messsage_idx += 1
+                if remove_think:
+                    if '<think>' in chunk['data']['text'] and not think_start:
+                        think_start = True
+                        continue
+                    if '</think>' in chunk['data']['text'] and not think_end:
+                        import re
+
+                        content = re.sub(r'^\n</think>', '', chunk['data']['text'])
+                        workflow_contents += content
+                        think_end = True
+                    elif think_end:
+                        workflow_contents += chunk['data']['text']
+                    if think_start:
+                        continue
+
+                else:
+                    workflow_contents += chunk['data']['text']
+
+            if chunk['event'] == 'node_started':
+                if chunk['data']['node_type'] == 'start' or chunk['data']['node_type'] == 'end':
+                    continue
+                messsage_idx += 1
+                msg = provider_message.MessageChunk(
+                    role='assistant',
+                    content=None,
+                    tool_calls=[
+                        provider_message.ToolCall(
+                            id=chunk['data']['node_id'],
+                            type='function',
+                            function=provider_message.FunctionCall(
+                                name=chunk['data']['title'],
+                                arguments=json.dumps({}),
+                            ),
+                        )
+                    ],
+                )
+
+                yield msg
+
+            if messsage_idx % 8 == 0 or is_final:
+                yield provider_message.MessageChunk(
+                    role='assistant',
+                    content=workflow_contents,
+                    is_final=is_final,
+                )
+
     async def run(self, query: pipeline_query.Query) -> typing.AsyncGenerator[provider_message.Message, None]:
         """运行请求"""
-        if self.pipeline_config['ai']['dify-service-api']['app-type'] == 'chat':
-            async for msg in self._chat_messages(query):
-                yield msg
-        elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'agent':
-            async for msg in self._agent_chat_messages(query):
-                yield msg
-        elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'workflow':
-            async for msg in self._workflow_messages(query):
-                yield msg
+        if await query.adapter.is_stream_output_supported():
+            msg_idx = 0
+            if self.pipeline_config['ai']['dify-service-api']['app-type'] == 'chat':
+                async for msg in self._chat_messages_chunk(query):
+                    msg_idx += 1
+                    msg.msg_sequence = msg_idx
+                    yield msg
+            elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'agent':
+                async for msg in self._agent_chat_messages_chunk(query):
+                    msg_idx += 1
+                    msg.msg_sequence = msg_idx
+                    yield msg
+            elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'workflow':
+                async for msg in self._workflow_messages_chunk(query):
+                    msg_idx += 1
+                    msg.msg_sequence = msg_idx
+                    yield msg
+            else:
+                raise errors.DifyAPIError(
+                    f'不支持的 Dify 应用类型: {self.pipeline_config["ai"]["dify-service-api"]["app-type"]}'
+                )
         else:
-            raise errors.DifyAPIError(
-                f'不支持的 Dify 应用类型: {self.pipeline_config["ai"]["dify-service-api"]["app-type"]}'
-            )
+            if self.pipeline_config['ai']['dify-service-api']['app-type'] == 'chat':
+                async for msg in self._chat_messages(query):
+                    yield msg
+            elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'agent':
+                async for msg in self._agent_chat_messages(query):
+                    yield msg
+            elif self.pipeline_config['ai']['dify-service-api']['app-type'] == 'workflow':
+                async for msg in self._workflow_messages(query):
+                    yield msg
+            else:
+                raise errors.DifyAPIError(
+                    f'不支持的 Dify 应用类型: {self.pipeline_config["ai"]["dify-service-api"]["app-type"]}'
+                )

pkg/provider/runners/localagent.py

@@ -1,39 +1,179 @@
 from __future__ import annotations
 
 import json
+import copy
 import typing
-
 from .. import runner
 import langbot_plugin.api.entities.builtin.pipeline.query as pipeline_query
 import langbot_plugin.api.entities.builtin.provider.message as provider_message
 
 
+rag_combined_prompt_template = """
+The following are relevant context entries retrieved from the knowledge base. 
+Please use them to answer the user's message. 
+Respond in the same language as the user's input.
+
+<context>
+{rag_context}
+</context>
+
+<user_message>
+{user_message}
+</user_message>
+"""
+
+
 @runner.runner_class('local-agent')
 class LocalAgentRunner(runner.RequestRunner):
     """本地Agent请求运行器"""
 
-    async def run(self, query: pipeline_query.Query) -> typing.AsyncGenerator[provider_message.Message, None]:
+    class ToolCallTracker:
+        """工具调用追踪器"""
+
+        def __init__(self):
+            self.active_calls: dict[str, dict] = {}
+            self.completed_calls: list[provider_message.ToolCall] = []
+
+    async def run(
+        self, query: pipeline_query.Query
+    ) -> typing.AsyncGenerator[provider_message.Message | provider_message.MessageChunk, None]:
         """运行请求"""
         pending_tool_calls = []
 
-        req_messages = query.prompt.messages.copy() + query.messages.copy() + [query.user_message]
+        kb_uuid = query.pipeline_config['ai']['local-agent']['knowledge-base']
 
-        use_llm_model = await self.ap.model_mgr.get_model_by_uuid(query.use_llm_model_uuid)
+        if kb_uuid == '__none__':
+            kb_uuid = None
 
-        # 首次请求
-        msg = await use_llm_model.requester.invoke_llm(
-            query,
-            use_llm_model,
-            req_messages,
-            query.use_funcs,
-            extra_args=use_llm_model.model_entity.extra_args,
-        )
+        user_message = copy.deepcopy(query.user_message)
 
-        yield msg
+        user_message_text = ''
 
-        pending_tool_calls = msg.tool_calls
+        if isinstance(user_message.content, str):
+            user_message_text = user_message.content
+        elif isinstance(user_message.content, list):
+            for ce in user_message.content:
+                if ce.type == 'text':
+                    user_message_text += ce.text
+                    break
 
-        req_messages.append(msg)
+        if kb_uuid and user_message_text:
+            # only support text for now
+            kb = await self.ap.rag_mgr.get_knowledge_base_by_uuid(kb_uuid)
+
+            if not kb:
+                self.ap.logger.warning(f'Knowledge base {kb_uuid} not found')
+                raise ValueError(f'Knowledge base {kb_uuid} not found')
+
+            result = await kb.retrieve(user_message_text, kb.knowledge_base_entity.top_k)
+
+            final_user_message_text = ''
+
+            if result:
+                rag_context = '\n\n'.join(
+                    f'[{i + 1}] {entry.metadata.get("text", "")}' for i, entry in enumerate(result)
+                )
+                final_user_message_text = rag_combined_prompt_template.format(
+                    rag_context=rag_context, user_message=user_message_text
+                )
+
+            else:
+                final_user_message_text = user_message_text
+
+            self.ap.logger.debug(f'Final user message text: {final_user_message_text}')
+
+            for ce in user_message.content:
+                if ce.type == 'text':
+                    ce.text = final_user_message_text
+                    break
+
+        req_messages = query.prompt.messages.copy() + query.messages.copy() + [user_message]
+
+        try:
+            is_stream = await query.adapter.is_stream_output_supported()
+        except AttributeError:
+            is_stream = False
+
+        remove_think = self.pipeline_config['output'].get('misc', '').get('remove-think')
+
+        if not is_stream:
+            # 非流式输出，直接请求
+
+            msg = await query.use_llm_model.requester.invoke_llm(
+                query,
+                query.use_llm_model,
+                req_messages,
+                query.use_funcs,
+                extra_args=query.use_llm_model.model_entity.extra_args,
+                remove_think=remove_think,
+            )
+            yield msg
+            final_msg = msg
+        else:
+            # 流式输出，需要处理工具调用
+            tool_calls_map: dict[str, provider_message.ToolCall] = {}
+            msg_idx = 0
+            accumulated_content = ''  # 从开始累积的所有内容
+            last_role = 'assistant'
+            msg_sequence = 1
+            async for msg in query.use_llm_model.requester.invoke_llm_stream(
+                query,
+                query.use_llm_model,
+                req_messages,
+                query.use_funcs,
+                extra_args=query.use_llm_model.model_entity.extra_args,
+                remove_think=remove_think,
+            ):
+                msg_idx = msg_idx + 1
+
+                # 记录角色
+                if msg.role:
+                    last_role = msg.role
+
+                # 累积内容
+                if msg.content:
+                    accumulated_content += msg.content
+
+                # 处理工具调用
+                if msg.tool_calls:
+                    for tool_call in msg.tool_calls:
+                        if tool_call.id not in tool_calls_map:
+                            tool_calls_map[tool_call.id] = provider_message.ToolCall(
+                                id=tool_call.id,
+                                type=tool_call.type,
+                                function=provider_message.FunctionCall(
+                                    name=tool_call.function.name if tool_call.function else '', arguments=''
+                                ),
+                            )
+                        if tool_call.function and tool_call.function.arguments:
+                            # 流式处理中，工具调用参数可能分多个chunk返回，需要追加而不是覆盖
+                            tool_calls_map[tool_call.id].function.arguments += tool_call.function.arguments
+                # continue
+                # 每8个chunk或最后一个chunk时，输出所有累积的内容
+                if msg_idx % 8 == 0 or msg.is_final:
+                    msg_sequence += 1
+                    yield provider_message.MessageChunk(
+                        role=last_role,
+                        content=accumulated_content,  # 输出所有累积内容
+                        tool_calls=list(tool_calls_map.values()) if (tool_calls_map and msg.is_final) else None,
+                        is_final=msg.is_final,
+                        msg_sequence=msg_sequence,
+                    )
+
+            # 创建最终消息用于后续处理
+            final_msg = provider_message.MessageChunk(
+                role=last_role,
+                content=accumulated_content,
+                tool_calls=list(tool_calls_map.values()) if tool_calls_map else None,
+                msg_sequence=msg_sequence,
+            )
+
+        pending_tool_calls = final_msg.tool_calls
+        first_content = final_msg.content
+        if isinstance(final_msg, provider_message.MessageChunk):
+            first_end_sequence = final_msg.msg_sequence
+
+        req_messages.append(final_msg)
 
         # 持续请求，只要还有待处理的工具调用就继续处理调用
         while pending_tool_calls:
@@ -43,13 +183,19 @@ class LocalAgentRunner(runner.RequestRunner):
 
                     parameters = json.loads(func.arguments)
 
-                    func_ret = await self.ap.tool_mgr.execute_func_call(func.name, parameters)
-
-                    msg = provider_message.Message(
-                        role='tool',
-                        content=json.dumps(func_ret, ensure_ascii=False),
-                        tool_call_id=tool_call.id,
-                    )
+                    func_ret = await self.ap.tool_mgr.execute_func_call(query, func.name, parameters)
+                    if is_stream:
+                        msg = provider_message.MessageChunk(
+                            role='tool',
+                            content=json.dumps(func_ret, ensure_ascii=False),
+                            tool_call_id=tool_call.id,
+                        )
+                    else:
+                        msg = provider_message.Message(
+                            role='tool',
+                            content=json.dumps(func_ret, ensure_ascii=False),
+                            tool_call_id=tool_call.id,
+                        )
 
                     yield msg
 
@@ -62,17 +208,81 @@ class LocalAgentRunner(runner.RequestRunner):
 
                     req_messages.append(err_msg)
 
-            # 处理完所有调用，再次请求
-            msg = await use_llm_model.requester.invoke_llm(
-                query,
-                use_llm_model,
-                req_messages,
-                query.use_funcs,
-                extra_args=use_llm_model.model_entity.extra_args,
-            )
+            if is_stream:
+                tool_calls_map = {}
+                msg_idx = 0
+                accumulated_content = ''  # 从开始累积的所有内容
+                last_role = 'assistant'
+                msg_sequence = first_end_sequence
 
-            yield msg
+                async for msg in query.use_llm_model.requester.invoke_llm_stream(
+                    query,
+                    query.use_llm_model,
+                    req_messages,
+                    query.use_funcs,
+                    extra_args=query.use_llm_model.model_entity.extra_args,
+                    remove_think=remove_think,
+                ):
+                    msg_idx += 1
 
-            pending_tool_calls = msg.tool_calls
+                    # 记录角色
+                    if msg.role:
+                        last_role = msg.role
 
-            req_messages.append(msg)
+                    # 第一次请求工具调用时的内容
+                    if msg_idx == 1:
+                        accumulated_content = first_content if first_content is not None else accumulated_content
+
+                    # 累积内容
+                    if msg.content:
+                        accumulated_content += msg.content
+
+                    # 处理工具调用
+                    if msg.tool_calls:
+                        for tool_call in msg.tool_calls:
+                            if tool_call.id not in tool_calls_map:
+                                tool_calls_map[tool_call.id] = provider_message.ToolCall(
+                                    id=tool_call.id,
+                                    type=tool_call.type,
+                                    function=provider_message.FunctionCall(
+                                        name=tool_call.function.name if tool_call.function else '', arguments=''
+                                    ),
+                                )
+                            if tool_call.function and tool_call.function.arguments:
+                                # 流式处理中，工具调用参数可能分多个chunk返回，需要追加而不是覆盖
+                                tool_calls_map[tool_call.id].function.arguments += tool_call.function.arguments
+
+                    # 每8个chunk或最后一个chunk时，输出所有累积的内容
+                    if msg_idx % 8 == 0 or msg.is_final:
+                        msg_sequence += 1
+                        yield provider_message.MessageChunk(
+                            role=last_role,
+                            content=accumulated_content,  # 输出所有累积内容
+                            tool_calls=list(tool_calls_map.values()) if (tool_calls_map and msg.is_final) else None,
+                            is_final=msg.is_final,
+                            msg_sequence=msg_sequence,
+                        )
+
+                final_msg = provider_message.MessageChunk(
+                    role=last_role,
+                    content=accumulated_content,
+                    tool_calls=list(tool_calls_map.values()) if tool_calls_map else None,
+                    msg_sequence=msg_sequence,
+                )
+            else:
+                # 处理完所有调用，再次请求
+                msg = await query.use_llm_model.requester.invoke_llm(
+                    query,
+                    query.use_llm_model,
+                    req_messages,
+                    query.use_funcs,
+                    extra_args=query.use_llm_model.model_entity.extra_args,
+                    remove_think=remove_think,
+                )
+
+                yield msg
+                final_msg = msg
+
+            pending_tool_calls = final_msg.tool_calls
+
+            req_messages.append(final_msg)