diff --git a/src/langbot/pkg/platform/adapters/lark/adapter.py b/src/langbot/pkg/platform/adapters/lark/adapter.py
index 7b32c4400..c6b677595 100644
--- a/src/langbot/pkg/platform/adapters/lark/adapter.py
+++ b/src/langbot/pkg/platform/adapters/lark/adapter.py
@@ -58,6 +58,8 @@ import lark_oapi.ws.exception
 import pydantic
 import quart
 
+from .stream_card import build_card, streaming_text
+
 import langbot_plugin.api.definition.abstract.platform.adapter as abstract_platform_adapter
 import langbot_plugin.api.definition.abstract.platform.event_logger as abstract_platform_logger
 from langbot.pkg.platform.adapters.lark.api_impl import LarkAPIMixin
@@ -490,28 +492,13 @@ class LarkAdapter(LarkAPIMixin, abstract_platform_adapter.AbstractPlatformAdapte
             self.card_sequence_dict.pop(old_card, None)
             self.card_last_update_dict.pop(old_card, None)
             self.closed_streaming_cards.discard(old_card)
-        card_data = {
-            'schema': '2.0',
-            'config': {
-                'update_multi': True,
-                'streaming_mode': True,
-                'streaming_config': {
-                    'print_step': {'default': 1},
-                    'print_frequency_ms': {'default': 70},
-                    'print_strategy': 'fast',
-                },
-            },
-            'body': {
-                'direction': 'vertical',
-                'elements': [{'tag': 'markdown', 'content': '', 'element_id': 'streaming_txt'}],
-            },
-        }
+        card_data = build_card(streaming=True)
         request = (
             CreateCardRequest.builder()
             .request_body(CreateCardRequestBody.builder().type('card_json').data(json.dumps(card_data)).build())
             .build()
         )
-        response: CreateCardResponse = self.api_client.cardkit.v1.card.create(request)
+        response: CreateCardResponse = await self.api_client.cardkit.v1.card.acreate(request)
         if not response.success():
             raise RuntimeError(f'Lark create_card failed: {response.code} {response.msg}')
         card_id = str(response.data.card_id)
@@ -531,16 +518,9 @@ class LarkAdapter(LarkAPIMixin, abstract_platform_adapter.AbstractPlatformAdapte
     def _streaming_mode_closed(response: ContentCardElementResponse) -> bool:
         return response.code == 300309 or 'streaming mode is closed' in str(response.msg).lower()
 
-    async def _replace_streaming_card(self, card_id: str, content: str) -> None:
+    async def _replace_streaming_card(self, card_id: str, content: str, *, finished: bool = True) -> None:
         sequence = self._next_card_sequence(card_id)
-        card_data = {
-            'schema': '2.0',
-            'config': {'update_multi': True},
-            'body': {
-                'direction': 'vertical',
-                'elements': [{'tag': 'markdown', 'content': content}],
-            },
-        }
+        card_data = build_card(content, finished=finished)
         request = (
             UpdateCardRequest.builder()
             .card_id(card_id)
@@ -583,14 +563,14 @@ class LarkAdapter(LarkAPIMixin, abstract_platform_adapter.AbstractPlatformAdapte
                 for paragraph in text_elements
             )
         if (is_final and not bot_message.tool_calls) or card_id in self.closed_streaming_cards:
-            await self._replace_streaming_card(card_id, content)
+            await self._replace_streaming_card(card_id, content, finished=is_final and not bot_message.tool_calls)
         else:
             sequence = self._next_card_sequence(card_id)
             request = (
                 ContentCardElementRequest.builder()
                 .card_id(card_id)
                 .element_id('streaming_txt')
-                .request_body(ContentCardElementRequestBody.builder().content(content).sequence(sequence).build())
+                .request_body(ContentCardElementRequestBody.builder().content(streaming_text(content)).sequence(sequence).build())
                 .build()
             )
             response: ContentCardElementResponse = await self.api_client.cardkit.v1.card_element.acontent(
@@ -598,7 +578,7 @@ class LarkAdapter(LarkAPIMixin, abstract_platform_adapter.AbstractPlatformAdapte
             )
             if not response.success():
                 if self._streaming_mode_closed(response):
-                    await self._replace_streaming_card(card_id, content)
+                    await self._replace_streaming_card(card_id, content, finished=is_final and not bot_message.tool_calls)
                 else:
                     raise RuntimeError(f'Lark card_element update failed: {response.code} {response.msg}')
         self.card_last_update_dict[card_id] = now
diff --git a/src/langbot/pkg/platform/adapters/lark/stream_card.py b/src/langbot/pkg/platform/adapters/lark/stream_card.py
new file mode 100644
index 000000000..88b165eef
--- /dev/null
+++ b/src/langbot/pkg/platform/adapters/lark/stream_card.py
@@ -0,0 +1,69 @@
+"""Compact streaming previews and a final card with collapsible reasoning."""
+
+import re
+
+
+def split_content(content: str) -> tuple[str, str]:
+    reasoning, answer = [], []
+    inside, offset = False, 0
+    for marker in re.finditer(r'</?think>', content, re.IGNORECASE):
+        (reasoning if inside else answer).append(content[offset:marker.start()])
+        inside = marker.group().lower() == '<think>'
+        offset = marker.end()
+    tail = content[offset:]
+    for size in range(min(7, len(tail)), 0, -1):
+        if any(tag.startswith(tail[-size:].lower()) for tag in ('<think>', '</think>')):
+            tail = tail[:-size]
+            break
+    (reasoning if inside else answer).append(tail)
+    return tuple('\n\n'.join(part.strip() for part in pieces if part.strip()) for pieces in (reasoning, answer))
+
+
+def streaming_text(content: str) -> str:
+    reasoning, answer = split_content(content)
+    if not reasoning:
+        return answer
+    # Keep live cards compact; the final card retains the complete reasoning.
+    preview = reasoning[-360:]
+    if len(reasoning) > 360:
+        preview = '…' + preview
+    quote = '> ' + preview.replace('\n', '\n> ')
+    return '**思考过程 / Thinking**\n\n' + quote + ('\n\n---\n\n' + answer if answer else '')
+
+
+def build_card(content: str = '', *, streaming: bool = False, finished: bool = True) -> dict:
+    reasoning, answer = split_content(content)
+    elements = []
+    if streaming:
+        elements.append({'tag': 'markdown', 'content': streaming_text(content), 'element_id': 'streaming_txt'})
+    else:
+        if reasoning:
+            elements.append({
+                'tag': 'collapsible_panel', 'expanded': not finished,
+                'header': {'title': {'tag': 'plain_text', 'content': '思考过程 / Thinking'}},
+                'elements': [{'tag': 'markdown', 'content': reasoning}],
+            })
+        if answer:
+            elements.append({'tag': 'markdown', 'content': answer})
+    elements.append({
+        'tag': 'column_set', 'horizontal_spacing': '8px',
+        'columns': [
+            {'tag': 'column', 'width': 'weighted', 'weight': 1, 'vertical_align': 'center',
+             'elements': [{'tag': 'markdown', 'content': '内容由 AI 生成 / AI-generated',
+                           'text_size': 'notation'}]},
+            {'tag': 'column', 'width': 'auto', 'vertical_align': 'center',
+             'elements': [{'tag': 'markdown', 'content': '[LangBot](https://langbot.app?utm_source=feishu&utm_medium=bot_card&utm_campaign=langbot)', 'text_size': 'notation',
+                           'text_align': 'right',
+                           'icon': {'tag': 'custom_icon',
+                                    'img_key': 'img_v3_02p3_05c65d5d-9bad-440a-a2fb-c89571bfd5bg'}}]},
+        ],
+    })
+    config = {'update_multi': True}
+    if streaming:
+        config.update(streaming_mode=True, streaming_config={
+            'print_step': {'default': 3}, 'print_frequency_ms': {'default': 40}, 'print_strategy': 'fast',
+        })
+    return {
+        'schema': '2.0', 'config': config,
+        'body': {'direction': 'vertical', 'padding': '12px 12px 12px 12px', 'elements': elements},
+    }
diff --git a/tests/unit_tests/platform/test_lark_eba_adapter.py b/tests/unit_tests/platform/test_lark_eba_adapter.py
index 9590fbed8..2dd23fd2b 100644
--- a/tests/unit_tests/platform/test_lark_eba_adapter.py
+++ b/tests/unit_tests/platform/test_lark_eba_adapter.py
@@ -705,7 +705,9 @@ async def test_lark_streaming_card_uses_strictly_increasing_sequences():
     final_request = adapter.api_client.cardkit.v1.card.aupdate.call_args.args[0]
     assert final_request.request_body.sequence == 3
     final_card = json.loads(final_request.request_body.card.data)
-    assert final_card['body']['elements'] == [{'tag': 'markdown', 'content': 'answer'}]
+    assert final_card['body']['elements'][0] == {'tag': 'markdown', 'content': 'answer'}
+    assert 'header' not in final_card
+    assert final_card['body']['elements'][-1]['columns'][1]['elements'][0]['content'] == '[LangBot](https://langbot.app?utm_source=feishu&utm_medium=bot_card&utm_campaign=langbot)'
     assert not final_card['config'].get('streaming_mode', False)
     assert 'response-1' not in adapter.card_id_dict
     assert 'stream-card-1' not in adapter.card_sequence_dict
@@ -910,3 +912,41 @@ async def test_lark_webhook_cardkit_submission_uses_async_card_update_only():
     response = await adapter.handle_unified_webhook('bot-1', '', request)
 
     assert response == {'toast': {'type': 'success', 'content': 'Submitted / 已提交'}}
+
+
+def test_lark_stream_card_separates_multiple_reasoning_rounds():
+    from langbot.pkg.platform.adapters.lark.stream_card import build_card, streaming_text
+    content = '<think>first thought</think>first answer<think>second thought</think>final answer'
+    preview = streaming_text(content)
+    assert '<think>' not in preview and '</think>' not in preview
+    card = build_card(content)
+    panel, answer, footer = card['body']['elements']
+    assert panel['tag'] == 'collapsible_panel'
+    assert panel['expanded'] is False
+    assert panel['elements'][0]['content'] == 'first thought\n\nsecond thought'
+    assert answer['content'] == 'first answer\n\nfinal answer'
+    assert 'AI' in footer['columns'][0]['elements'][0]['content']
+    assert footer['columns'][1]['elements'][0]['icon']['tag'] == 'custom_icon'
+
+
+def test_lark_live_preview_is_bounded_but_final_reasoning_is_complete():
+    from langbot.pkg.platform.adapters.lark.stream_card import build_card, streaming_text
+    thought = 'reasoning ' * 200
+    content = '<think>' + thought + '</thi'
+    assert len(streaming_text(content)) < 450
+    assert '</thi' not in streaming_text(content)
+    card = build_card(content)
+    assert card['body']['elements'][0]['elements'][0]['content'] == thought.strip()
+    assert build_card('plain answer')['body']['elements'][0]['content'] == 'plain answer'
+
+
+@pytest.mark.asyncio
+async def test_lark_stream_card_creation_is_async_and_branded():
+    adapter = make_adapter()
+    await adapter.create_card_id('response-1')
+    request = adapter.api_client.cardkit.v1.card.acreate.await_args.args[0]
+    card = json.loads(request.request_body.data)
+    assert 'header' not in card
+    assert card['body']['elements'][-1]['columns'][1]['elements'][0]['content'] == '[LangBot](https://langbot.app?utm_source=feishu&utm_medium=bot_card&utm_campaign=langbot)'
+    assert card['config']['streaming_mode'] is True
+    assert card['body']['elements'][0]['element_id'] == 'streaming_txt'