代码提交

2026-06-08 22:52:18 +08:00
1 parent b101ee7fc9
commit 1938c80061
@@ -4,7 +4,8 @@
      "PowerShell(uv --version)",
      "PowerShell(netstat -ano)",
      "PowerShell(Get-NetTCPConnection -LocalPort 8001 -ErrorAction SilentlyContinue)",
-      "PowerShell(Get-Process -Id 20560 -ErrorAction SilentlyContinue | Select-Object Id,ProcessName,Path,StartTime; try { \\(Invoke-WebRequest -Uri \"http://127.0.0.1:8001/health\" -UseBasicParsing -TimeoutSec 5\\).Content } catch { \"health check failed: $_\" })"
+      "PowerShell(Get-Process -Id 20560 -ErrorAction SilentlyContinue | Select-Object Id,ProcessName,Path,StartTime; try { \\(Invoke-WebRequest -Uri \"http://127.0.0.1:8001/health\" -UseBasicParsing -TimeoutSec 5\\).Content } catch { \"health check failed: $_\" })",
+      "Bash(curl -s -X POST -H \"Content-Type: application/json\" -d '{\"username\":\"demo\"}' http://localhost:8001/api/v1/auth/login/username)"
    ]
  }
 }
@@ -17,7 +17,7 @@ SERPER_API_KEY=your-serper-api-key
 # AUTH_JWT_SECRET must stay stable across restarts, or existing logins expire.
 # GATEWAY_CORS_ORIGINS must include the exact frontend origins, separated by commas.
 AUTH_JWT_SECRET=i3_BiBK44sYHiFIkChhDs46y57ZewJKO3diQDw01anI
-GATEWAY_CORS_ORIGINS=http://localhost:5173,http://127.0.0.1:5173,http://localhost:5174,http://127.0.0.1:5174,http://localhost:4200
+GATEWAY_CORS_ORIGINS=http://localhost:5173,http://127.0.0.1:5173,http://localhost:5183,http://localhost:5174,http://127.0.0.1:5174,http://localhost:4200
 GATEWAY_CORS_ALLOW_ALL=0
 GATEWAY_ENABLE_DOCS=true
 GATEWAY_WORKERS=2
@@ -0,0 +1,760 @@
+# This file was autogenerated by uv via the following command:
+#    uv export --no-hashes --no-emit-project --no-dev -o _reqs.txt
+-e ./packages/harness
+    # via deer-flow
+agent-client-protocol==0.10.1
+    # via deerflow-harness
+agent-sandbox==0.0.30
+    # via deerflow-harness
+aiohappyeyeballs==2.6.2
+    # via aiohttp
+aiohttp==3.14.1
+    # via
+    #   aiohttp-retry
+    #   dingtalk-stream
+    #   firecrawl-py
+    #   hindsight-client
+    #   kubernetes
+    #   wecom-aibot-python-sdk
+aiohttp-retry==2.9.1
+    # via hindsight-client
+aiosignal==1.4.0
+    # via aiohttp
+aiosqlite==0.22.1
+    # via
+    #   deerflow-harness
+    #   langgraph-checkpoint-sqlite
+alembic==1.18.4
+    # via deerflow-harness
+annotated-doc==0.0.4
+    # via fastapi
+annotated-types==0.7.0
+    # via pydantic
+anthropic==0.107.1
+    # via langchain-anthropic
+anyio==4.13.0
+    # via
+    #   anthropic
+    #   google-genai
+    #   httpx
+    #   mcp
+    #   openai
+    #   sse-starlette
+    #   starlette
+    #   watchfiles
+asyncmy==0.2.11
+    # via deer-flow
+attrs==26.1.0
+    # via
+    #   aiohttp
+    #   jsonschema
+    #   referencing
+audioop-lts==0.2.2 ; python_full_version >= '3.13'
+    # via
+    #   speechrecognition
+    #   standard-aifc
+azure-ai-documentintelligence==1.0.2
+    # via markitdown
+azure-core==1.41.0
+    # via
+    #   azure-ai-documentintelligence
+    #   azure-identity
+azure-identity==1.25.3
+    # via markitdown
+backoff==2.2.1
+    # via langfuse
+bcrypt==5.0.0
+    # via deer-flow
+beautifulsoup4==4.15.0
+    # via
+    #   markdownify
+    #   markitdown
+    #   readabilipy
+blockbuster==1.5.26
+    # via langgraph-runtime-inmem
+brotli==1.2.0 ; platform_python_implementation == 'CPython'
+    # via httpx
+brotlicffi==1.2.0.1 ; platform_python_implementation != 'CPython'
+    # via httpx
+certifi==2026.5.20
+    # via
+    #   httpcore
+    #   httpx
+    #   kubernetes
+    #   requests
+    #   volcengine-python-sdk
+    #   wecom-aibot-python-sdk
+cffi==2.0.0
+    # via
+    #   brotlicffi
+    #   cryptography
+charset-normalizer==3.4.7
+    # via
+    #   markitdown
+    #   pdfminer-six
+    #   requests
+click==8.4.1
+    # via
+    #   ddgs
+    #   langgraph-cli
+    #   magika
+    #   uvicorn
+cloudpickle==3.1.2
+    # via langgraph-api
+cobble==0.1.4
+    # via mammoth
+colorama==0.4.6 ; sys_platform == 'win32'
+    # via
+    #   click
+    #   tqdm
+    #   uvicorn
+coloredlogs==15.0.1
+    # via onnxruntime
+croniter==6.2.2
+    # via
+    #   deer-flow
+    #   langgraph-runtime-inmem
+cryptography==48.0.0
+    # via
+    #   azure-identity
+    #   google-auth
+    #   langgraph-api
+    #   msal
+    #   pdfminer-six
+    #   pyjwt
+    #   wecom-aibot-python-sdk
+ddgs==9.14.4
+    # via deerflow-harness
+defusedxml==0.7.1
+    # via
+    #   markitdown
+    #   youtube-transcript-api
+dingtalk-stream==0.24.3
+    # via deer-flow
+distro==1.9.0
+    # via
+    #   anthropic
+    #   google-genai
+    #   openai
+dnspython==2.8.0
+    # via email-validator
+docstring-parser==0.18.0
+    # via anthropic
+dotenv==0.9.9
+    # via deerflow-harness
+duckdb==1.5.3
+    # via deerflow-harness
+durationpy==0.10
+    # via kubernetes
+email-validator==2.3.0
+    # via deer-flow
+et-xmlfile==2.0.0
+    # via openpyxl
+exa-py==2.13.0
+    # via deerflow-harness
+fake-useragent==2.2.0
+    # via ddgs
+fastapi==0.136.3
+    # via deer-flow
+filetype==1.2.0
+    # via langchain-google-genai
+firecrawl-py==4.28.2
+    # via deerflow-harness
+flatbuffers==25.12.19
+    # via onnxruntime
+forbiddenfruit==0.1.4 ; implementation_name == 'cpython'
+    # via blockbuster
+frozenlist==1.8.0
+    # via
+    #   aiohttp
+    #   aiosignal
+google-auth==2.53.0
+    # via google-genai
+google-genai==2.8.0
+    # via langchain-google-genai
+googleapis-common-protos==1.75.0
+    # via opentelemetry-exporter-otlp-proto-http
+greenlet==3.5.1
+    # via sqlalchemy
+grpcio==1.80.0
+    # via
+    #   grpcio-health-checking
+    #   grpcio-tools
+    #   langgraph-api
+grpcio-health-checking==1.80.0
+    # via langgraph-api
+grpcio-tools==1.80.0
+    # via langgraph-api
+h11==0.16.0
+    # via
+    #   httpcore
+    #   uvicorn
+h2==4.3.0
+    # via httpx
+hindsight-client==0.7.2
+    # via deer-flow
+hpack==4.1.0
+    # via h2
+html5lib==1.1
+    # via readabilipy
+httpcore==1.0.9
+    # via
+    #   exa-py
+    #   httpx
+    #   python-telegram-bot
+httptools==0.8.0
+    # via
+    #   langgraph-api
+    #   uvicorn
+httpx==0.28.1
+    # via
+    #   agent-sandbox
+    #   anthropic
+    #   ddgs
+    #   deer-flow
+    #   deerflow-harness
+    #   exa-py
+    #   firecrawl-py
+    #   google-genai
+    #   langfuse
+    #   langgraph-api
+    #   langgraph-cli
+    #   langgraph-sdk
+    #   langsmith
+    #   lark-oapi
+    #   mcp
+    #   openai
+    #   python-telegram-bot
+    #   tavily-python
+httpx-sse==0.4.3
+    # via mcp
+humanfriendly==10.0
+    # via coloredlogs
+hyperframe==6.1.0
+    # via h2
+idna==3.18
+    # via
+    #   anyio
+    #   email-validator
+    #   httpx
+    #   requests
+    #   yarl
+isodate==0.7.2
+    # via azure-ai-documentintelligence
+jiter==0.15.0
+    # via
+    #   anthropic
+    #   openai
+jsonpatch==1.33
+    # via langchain-core
+jsonpointer==3.1.1
+    # via jsonpatch
+jsonschema==4.26.0
+    # via mcp
+jsonschema-rs==0.44.1
+    # via langgraph-api
+jsonschema-specifications==2025.9.1
+    # via jsonschema
+kubernetes==36.0.2
+    # via deerflow-harness
+langchain==1.3.4
+    # via deerflow-harness
+langchain-anthropic==1.4.4
+    # via deerflow-harness
+langchain-core==1.4.1
+    # via
+    #   langchain
+    #   langchain-anthropic
+    #   langchain-deepseek
+    #   langchain-google-genai
+    #   langchain-mcp-adapters
+    #   langchain-openai
+    #   langgraph
+    #   langgraph-api
+    #   langgraph-checkpoint
+    #   langgraph-prebuilt
+    #   langgraph-sdk
+langchain-deepseek==1.1.0
+    # via deerflow-harness
+langchain-google-genai==4.2.4
+    # via deerflow-harness
+langchain-mcp-adapters==0.2.2
+    # via deerflow-harness
+langchain-openai==1.2.2
+    # via
+    #   deerflow-harness
+    #   langchain-deepseek
+langchain-protocol==0.0.16
+    # via
+    #   langchain-core
+    #   langgraph-api
+    #   langgraph-sdk
+langfuse==4.7.1
+    # via deerflow-harness
+langgraph==1.2.4
+    # via
+    #   deerflow-harness
+    #   langchain
+    #   langgraph-api
+    #   langgraph-runtime-inmem
+langgraph-api==0.9.0
+    # via deerflow-harness
+langgraph-checkpoint==4.1.1
+    # via
+    #   langgraph
+    #   langgraph-api
+    #   langgraph-checkpoint-postgres
+    #   langgraph-checkpoint-sqlite
+    #   langgraph-prebuilt
+    #   langgraph-runtime-inmem
+langgraph-checkpoint-postgres==3.1.0
+    # via deerflow-harness
+langgraph-checkpoint-sqlite==3.1.0
+    # via deerflow-harness
+langgraph-cli==0.4.27
+    # via deerflow-harness
+langgraph-prebuilt==1.1.0
+    # via langgraph
+langgraph-runtime-inmem==0.29.0
+    # via
+    #   deerflow-harness
+    #   langgraph-api
+langgraph-sdk==0.4.2
+    # via
+    #   deer-flow
+    #   deerflow-harness
+    #   langgraph
+    #   langgraph-api
+    #   langgraph-cli
+langsmith==0.8.9
+    # via
+    #   langchain-core
+    #   langgraph-api
+lark-oapi==1.6.8
+    # via deer-flow
+lxml==6.1.1
+    # via
+    #   ddgs
+    #   markitdown
+    #   python-pptx
+    #   readabilipy
+magika==0.6.3
+    # via markitdown
+mako==1.3.12
+    # via alembic
+mammoth==1.11.0
+    # via markitdown
+markdown-to-mrkdwn==0.3.3
+    # via deer-flow
+markdownify==1.2.2
+    # via
+    #   deerflow-harness
+    #   markitdown
+markitdown==0.1.5
+    # via deerflow-harness
+markupsafe==3.0.3
+    # via mako
+mcp==1.27.2
+    # via langchain-mcp-adapters
+mpmath==1.3.0
+    # via sympy
+msal==1.37.0
+    # via
+    #   azure-identity
+    #   msal-extensions
+msal-extensions==1.3.1
+    # via azure-identity
+multidict==6.7.1
+    # via
+    #   aiohttp
+    #   yarl
+nest-asyncio==1.6.0
+    # via firecrawl-py
+numpy==2.4.6
+    # via
+    #   magika
+    #   onnxruntime
+    #   pandas
+oauthlib==3.3.1
+    # via requests-oauthlib
+olefile==0.47
+    # via markitdown
+onnxruntime==1.20.1
+    # via magika
+openai==2.41.0
+    # via
+    #   exa-py
+    #   langchain-openai
+openpyxl==3.1.5
+    # via markitdown
+opentelemetry-api==1.42.1
+    # via
+    #   langfuse
+    #   langgraph-api
+    #   langsmith
+    #   opentelemetry-exporter-otlp-proto-http
+    #   opentelemetry-sdk
+    #   opentelemetry-semantic-conventions
+opentelemetry-exporter-otlp-proto-common==1.42.1
+    # via opentelemetry-exporter-otlp-proto-http
+opentelemetry-exporter-otlp-proto-http==1.42.1
+    # via
+    #   langfuse
+    #   langgraph-api
+    #   langsmith
+opentelemetry-proto==1.42.1
+    # via
+    #   opentelemetry-exporter-otlp-proto-common
+    #   opentelemetry-exporter-otlp-proto-http
+opentelemetry-sdk==1.42.1
+    # via
+    #   langfuse
+    #   langgraph-api
+    #   langsmith
+    #   opentelemetry-exporter-otlp-proto-http
+opentelemetry-semantic-conventions==0.63b1
+    # via opentelemetry-sdk
+orjson==3.11.9
+    # via
+    #   langgraph-api
+    #   langgraph-checkpoint-postgres
+    #   langgraph-sdk
+    #   langsmith
+ormsgpack==1.12.2
+    # via langgraph-checkpoint
+packaging==26.2
+    # via
+    #   langchain-core
+    #   langfuse
+    #   langsmith
+    #   onnxruntime
+pandas==3.0.3
+    # via markitdown
+pathspec==1.1.1
+    # via langgraph-cli
+pdfminer-six==20251230
+    # via
+    #   markitdown
+    #   pdfplumber
+pdfplumber==0.11.9
+    # via markitdown
+pillow==12.2.0
+    # via
+    #   pdfplumber
+    #   python-pptx
+primp==1.3.1
+    # via ddgs
+propcache==0.5.2
+    # via
+    #   aiohttp
+    #   yarl
+protobuf==6.33.6
+    # via
+    #   googleapis-common-protos
+    #   grpcio-health-checking
+    #   grpcio-tools
+    #   langgraph-api
+    #   onnxruntime
+    #   opentelemetry-proto
+psycopg==3.3.4
+    # via
+    #   deerflow-harness
+    #   langgraph-checkpoint-postgres
+psycopg-binary==3.3.4 ; implementation_name != 'pypy'
+    # via psycopg
+psycopg-pool==3.3.1
+    # via
+    #   deerflow-harness
+    #   langgraph-checkpoint-postgres
+pyasn1==0.6.3
+    # via pyasn1-modules
+pyasn1-modules==0.4.2
+    # via google-auth
+pycparser==3.0 ; implementation_name != 'PyPy'
+    # via cffi
+pycryptodome==3.23.0
+    # via lark-oapi
+pydantic==2.13.4
+    # via
+    #   agent-client-protocol
+    #   agent-sandbox
+    #   anthropic
+    #   deerflow-harness
+    #   exa-py
+    #   fastapi
+    #   firecrawl-py
+    #   google-genai
+    #   hindsight-client
+    #   langchain
+    #   langchain-anthropic
+    #   langchain-core
+    #   langchain-google-genai
+    #   langfuse
+    #   langgraph
+    #   langsmith
+    #   mcp
+    #   openai
+    #   pydantic-settings
+pydantic-core==2.46.4
+    # via pydantic
+pydantic-settings==2.14.1
+    # via mcp
+pydub==0.25.1
+    # via markitdown
+pyee==13.0.1
+    # via wecom-aibot-python-sdk
+pyjwt==2.13.0
+    # via
+    #   deer-flow
+    #   langgraph-api
+    #   mcp
+    #   msal
+pypdfium2==5.9.0
+    # via pdfplumber
+pyreadline3==3.5.6 ; sys_platform == 'win32'
+    # via humanfriendly
+python-dateutil==2.9.0.post0
+    # via
+    #   croniter
+    #   hindsight-client
+    #   kubernetes
+    #   pandas
+    #   volcengine-python-sdk
+python-dotenv==1.2.2
+    # via
+    #   dotenv
+    #   exa-py
+    #   firecrawl-py
+    #   langgraph-cli
+    #   magika
+    #   pydantic-settings
+    #   uvicorn
+python-multipart==0.0.32
+    # via
+    #   deer-flow
+    #   mcp
+python-pptx==1.0.2
+    # via markitdown
+python-telegram-bot==22.7
+    # via deer-flow
+pywin32==312 ; sys_platform == 'win32'
+    # via mcp
+pyyaml==6.0.3
+    # via
+    #   deerflow-harness
+    #   kubernetes
+    #   langchain-core
+    #   uvicorn
+readabilipy==0.3.0
+    # via deerflow-harness
+referencing==0.37.0
+    # via
+    #   jsonschema
+    #   jsonschema-specifications
+regex==2026.5.9
+    # via
+    #   readabilipy
+    #   tiktoken
+requests==2.34.2
+    # via
+    #   azure-core
+    #   dingtalk-stream
+    #   exa-py
+    #   firecrawl-py
+    #   google-auth
+    #   google-genai
+    #   kubernetes
+    #   langsmith
+    #   lark-oapi
+    #   markitdown
+    #   msal
+    #   opentelemetry-exporter-otlp-proto-http
+    #   requests-oauthlib
+    #   requests-toolbelt
+    #   tavily-python
+    #   tiktoken
+    #   youtube-transcript-api
+requests-oauthlib==2.0.0
+    # via kubernetes
+requests-toolbelt==1.0.0
+    # via
+    #   langsmith
+    #   lark-oapi
+rpds-py==2026.5.1
+    # via
+    #   jsonschema
+    #   referencing
+setuptools==82.0.1
+    # via grpcio-tools
+six==1.17.0
+    # via
+    #   html5lib
+    #   kubernetes
+    #   markdownify
+    #   python-dateutil
+    #   volcengine-python-sdk
+slack-sdk==3.42.0
+    # via deer-flow
+sniffio==1.3.1
+    # via
+    #   anthropic
+    #   google-genai
+    #   openai
+socksio==1.0.0
+    # via httpx
+soupsieve==2.8.4
+    # via beautifulsoup4
+speechrecognition==3.16.1
+    # via markitdown
+sqlalchemy==2.0.50
+    # via
+    #   alembic
+    #   deerflow-harness
+sqlite-vec==0.1.9
+    # via langgraph-checkpoint-sqlite
+sse-starlette==3.3.4
+    # via
+    #   deer-flow
+    #   langgraph-api
+    #   langgraph-runtime-inmem
+    #   mcp
+standard-aifc==3.13.0 ; python_full_version >= '3.13'
+    # via speechrecognition
+standard-chunk==3.13.0 ; python_full_version >= '3.13'
+    # via standard-aifc
+starlette==1.2.1
+    # via
+    #   fastapi
+    #   langgraph-api
+    #   langgraph-runtime-inmem
+    #   mcp
+    #   sse-starlette
+structlog==25.5.0
+    # via
+    #   langgraph-api
+    #   langgraph-runtime-inmem
+sympy==1.14.0
+    # via onnxruntime
+tavily-python==0.7.25
+    # via deerflow-harness
+tenacity==9.1.4
+    # via
+    #   google-genai
+    #   langchain-core
+    #   langgraph-api
+tiktoken==0.13.0
+    # via
+    #   deerflow-harness
+    #   langchain-openai
+    #   tavily-python
+tqdm==4.68.1
+    # via openai
+truststore==0.10.4
+    # via langgraph-api
+typing-extensions==4.15.0
+    # via
+    #   aiohttp
+    #   aiosignal
+    #   alembic
+    #   anthropic
+    #   anyio
+    #   azure-ai-documentintelligence
+    #   azure-core
+    #   azure-identity
+    #   beautifulsoup4
+    #   exa-py
+    #   fastapi
+    #   google-genai
+    #   grpcio
+    #   hindsight-client
+    #   langchain-core
+    #   langchain-mcp-adapters
+    #   langchain-protocol
+    #   mcp
+    #   openai
+    #   opentelemetry-api
+    #   opentelemetry-exporter-otlp-proto-http
+    #   opentelemetry-sdk
+    #   opentelemetry-semantic-conventions
+    #   psycopg
+    #   psycopg-pool
+    #   pydantic
+    #   pydantic-core
+    #   pyee
+    #   python-pptx
+    #   referencing
+    #   speechrecognition
+    #   sqlalchemy
+    #   starlette
+    #   typing-inspection
+typing-inspection==0.4.2
+    # via
+    #   fastapi
+    #   mcp
+    #   pydantic
+    #   pydantic-settings
+tzdata==2026.2 ; sys_platform == 'emscripten' or sys_platform == 'win32'
+    # via
+    #   pandas
+    #   psycopg
+urllib3==2.7.0
+    # via
+    #   hindsight-client
+    #   kubernetes
+    #   requests
+    #   volcengine-python-sdk
+uuid-utils==0.16.0
+    # via
+    #   langchain-core
+    #   langgraph-api
+    #   langsmith
+uvicorn==0.49.0
+    # via
+    #   deer-flow
+    #   langgraph-api
+    #   mcp
+uvloop==0.22.1 ; sys_platform != 'win32'
+    # via
+    #   langgraph-api
+    #   uvicorn
+volcengine-python-sdk==5.0.33
+    # via agent-sandbox
+watchfiles==1.2.0
+    # via
+    #   langgraph-api
+    #   uvicorn
+webencodings==0.5.1
+    # via html5lib
+websocket-client==1.9.0
+    # via kubernetes
+websockets==15.0.1
+    # via
+    #   dingtalk-stream
+    #   firecrawl-py
+    #   google-genai
+    #   langgraph-api
+    #   langgraph-sdk
+    #   langsmith
+    #   lark-oapi
+    #   uvicorn
+    #   wecom-aibot-python-sdk
+wecom-aibot-python-sdk==1.0.2
+    # via deer-flow
+wrapt==1.17.3
+    # via langfuse
+xlrd==2.0.2
+    # via markitdown
+xlsxwriter==3.2.9
+    # via python-pptx
+xxhash==3.7.0
+    # via
+    #   langgraph
+    #   langsmith
+yarl==1.24.2
+    # via aiohttp
+youtube-transcript-api==1.0.3
+    # via markitdown
+zstandard==0.25.0
+    # via
+    #   langgraph-api
+    #   langsmith
@@ -30,9 +30,9 @@ from app.gateway.routers import (
    artifacts,
    assistants_compat,
    auth,
-    channels,
-    checkpoint_migration,
-    feedback,
+    channels,
+    checkpoint_migration,
+    feedback,
    html_page_favorites,
    intent,
    llm_metrics,
@@ -41,6 +41,8 @@ from app.gateway.routers import (
    models,
    multi_agent,
    notifications,
+    parallel_agent_tasks,
+    parallel_agents,
    public_embed,
    public_scheduled_tasks,
    recommend,
@@ -378,7 +380,7 @@ def _release_singleton_lock(handle) -> None:
            pass


-async def _curator_scheduler_loop() -> None:
+async def _curator_scheduler_loop() -> None:
    """常驻后台任务:每隔 _CURATOR_CHECK_INTERVAL_SECONDS 检查一次需不需要进化。

    多 worker 部署下用文件锁做 leader 选举 —— 只有抢到锁的那个 worker 进程真正
@@ -404,248 +406,248 @@ async def _curator_scheduler_loop() -> None:
            except Exception:
                logger.exception("Curator scheduler tick failed (non-fatal)")
            await asyncio.sleep(_CURATOR_CHECK_INTERVAL_SECONDS)
-    finally:
-        _release_singleton_lock(lock_handle)
-
-
-def _env_bool(name: str, default: bool) -> bool:
-    raw = os.getenv(name)
-    if raw is None:
-        return default
-    return raw.strip().lower() not in {"0", "false", "no", "off"}
-
-
-def _env_int(name: str, default: int, *, alias: str | None = None, minimum: int = 1) -> int:
-    raw = os.getenv(name)
-    if raw is None and alias:
-        raw = os.getenv(alias)
-    if raw is None:
-        raw = str(default)
-    try:
-        value = int(raw)
-    except ValueError:
-        return default
-    return max(minimum, value)
-
-
-def _env_float(name: str, default: float, *, alias: str | None = None, minimum: float = 0.0) -> float:
-    raw = os.getenv(name)
-    if raw is None and alias:
-        raw = os.getenv(alias)
-    if raw is None:
-        raw = str(default)
-    try:
-        value = float(raw)
-    except ValueError:
-        return default
-    return max(minimum, value)
-
-
-_LEADERBOARD_SNAPSHOT_BACKGROUND_ENABLED = _env_bool("ADMIN_LEADERBOARD_BACKGROUND_ENABLED", True)
-_LEADERBOARD_SNAPSHOT_PREWARM_DAYS = _env_int(
-    "ADMIN_LEADERBOARD_BACKGROUND_PREWARM_DAYS",
-    30,
-)
-_LEADERBOARD_SNAPSHOT_BATCH_SIZE = _env_int(
-    "ADMIN_LEADERBOARD_BACKGROUND_BATCH_SIZE",
-    1,
-)
-_LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS = _env_float(
-    "ADMIN_LEADERBOARD_BACKGROUND_INTERVAL_SECONDS",
-    300.0,
-)
-_LEADERBOARD_SNAPSHOT_SLEEP_SECONDS = _env_float(
-    "ADMIN_LEADERBOARD_BACKGROUND_SLEEP_SECONDS",
-    2.0,
-)
-_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS = _env_int(
-    "ADMIN_LEADERBOARD_BACKGROUND_RUNNING_TTL_SECONDS",
-    1800,
-)
-_LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS = _env_float(
-    "ADMIN_LEADERBOARD_BACKGROUND_STARTUP_DELAY_SECONDS",
-    60.0,
-)
-
-
-def _recent_leaderboard_stat_dates(days: int) -> list[str]:
-    from datetime import datetime, timedelta
-
-    from deerflow.persistence.types import BEIJING_TZ
-
-    today = datetime.now(BEIJING_TZ).date()
-    start = today - timedelta(days=max(1, days) - 1)
-    return [(start + timedelta(days=i)).isoformat() for i in range((today - start).days + 1)]
-
-
-async def _leaderboard_snapshot_due_dates(
-    session_factory,
-    *,
-    settings_hash: str,
-    limit: int,
-    running_ttl_seconds: int,
-) -> list[str]:
-    from datetime import UTC, datetime, timedelta
-
-    from sqlalchemy import and_, or_, select
-
-    from deerflow.persistence.admin_stats.model import AdminLeaderboardDailyStatRow
-
-    cutoff = datetime.now(UTC) - timedelta(seconds=running_ttl_seconds)
-    async with session_factory() as session:
-        rows = (
-            await session.execute(
-                select(AdminLeaderboardDailyStatRow.stat_date)
-                .where(
-                    AdminLeaderboardDailyStatRow.settings_hash == settings_hash,
-                    or_(
-                        AdminLeaderboardDailyStatRow.status.in_(["queued", "error"]),
-                        and_(
-                            AdminLeaderboardDailyStatRow.status == "running",
-                            AdminLeaderboardDailyStatRow.updated_at < cutoff,
-                        ),
-                    ),
-                )
-                .order_by(AdminLeaderboardDailyStatRow.stat_date.desc())
-                .limit(limit)
-            )
-        ).all()
-    return [str(row[0]) for row in rows]
-
-
-async def _claim_leaderboard_snapshot(
-    session_factory,
-    *,
-    stat_date: str,
-    settings_hash: str,
-    running_ttl_seconds: int,
-) -> bool:
-    from datetime import UTC, datetime
-
-    from app.gateway.routers.admin_users import _snapshot_is_fresh
-    from deerflow.persistence.admin_stats.model import AdminLeaderboardDailyStatRow
-    from deerflow.persistence.types import BEIJING_TZ
-
-    async with session_factory() as session:
-        row = await session.get(
-            AdminLeaderboardDailyStatRow,
-            {"stat_date": stat_date, "settings_hash": settings_hash},
-        )
-        if row is None or _snapshot_is_fresh(row, stat_date):
-            return False
-        if row.status == "running":
-            updated_at = row.updated_at
-            if isinstance(updated_at, datetime):
-                if updated_at.tzinfo is None:
-                    updated_at = updated_at.replace(tzinfo=BEIJING_TZ)
-                age = (datetime.now(BEIJING_TZ) - updated_at.astimezone(BEIJING_TZ)).total_seconds()
-                if age < running_ttl_seconds:
-                    return False
-        row.status = "running"
-        row.message = "排行榜后台统计任务已认领。"
-        row.error = None
-        row.updated_at = datetime.now(UTC)
-        await session.commit()
-    return True
-
-
-async def _run_leaderboard_snapshot_tick() -> tuple[int, int]:
-    from app.gateway.routers.admin_users import (
-        _daily_job_key,
-        _queue_daily_leaderboard_snapshots,
-        _refresh_daily_leaderboard_snapshot,
-        _resolve_leaderboard_snapshot_scope,
-    )
-    from deerflow.config.system_settings import load_system_settings
-    from deerflow.persistence.engine import get_session_factory
-
-    session_factory = get_session_factory()
-    if session_factory is None:
-        return 0, 0
-
-    settings = load_system_settings().leaderboard
-    excluded, include_scheduled, include_admins, include_failed, settings_hash = _resolve_leaderboard_snapshot_scope(settings)
-    stat_dates = _recent_leaderboard_stat_dates(_LEADERBOARD_SNAPSHOT_PREWARM_DAYS)
-    queued_dates = await _queue_daily_leaderboard_snapshots(
-        session_factory,
-        stat_dates,
-        settings_hash,
-        message="排行榜后台统计任务定期预热/补算队列。",
-    )
-
-    due_dates = await _leaderboard_snapshot_due_dates(
-        session_factory,
-        settings_hash=settings_hash,
-        limit=_LEADERBOARD_SNAPSHOT_BATCH_SIZE,
-        running_ttl_seconds=_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS,
-    )
-    processed = 0
-    for stat_date in due_dates:
-        claimed = await _claim_leaderboard_snapshot(
-            session_factory,
-            stat_date=stat_date,
-            settings_hash=settings_hash,
-            running_ttl_seconds=_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS,
-        )
-        if not claimed:
-            continue
-        logger.info("Leaderboard snapshot scheduler: building date=%s hash=%s", stat_date, settings_hash[:10])
-        await _refresh_daily_leaderboard_snapshot(
-            _daily_job_key(stat_date, settings_hash),
-            session_factory=session_factory,
-            settings=settings,
-            stat_date=stat_date,
-            settings_hash=settings_hash,
-            excluded=excluded,
-            include_scheduled=include_scheduled,
-            include_admins=include_admins,
-            include_failed=include_failed,
-        )
-        processed += 1
-        if _LEADERBOARD_SNAPSHOT_SLEEP_SECONDS > 0 and processed < len(due_dates):
-            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_SLEEP_SECONDS)
-    return processed, len(queued_dates)
-
-
-async def _leaderboard_snapshot_scheduler_loop() -> None:
-    """Background task inside the existing gateway process for daily snapshots."""
-    if not _LEADERBOARD_SNAPSHOT_BACKGROUND_ENABLED:
-        logger.info("Leaderboard snapshot scheduler disabled by ADMIN_LEADERBOARD_BACKGROUND_ENABLED")
-        return
-
-    from deerflow.config.runtime_paths import runtime_home
-
-    lock_handle = _acquire_singleton_lock(runtime_home() / ".leaderboard_snapshot_scheduler.lock")
-    if lock_handle is None:
-        logger.info("Leaderboard snapshot scheduler: another gateway worker holds the lock; this worker skips")
-        return
-
-    logger.info(
-        "Leaderboard snapshot scheduler started: prewarm_days=%d batch_size=%d interval=%.1fs",
-        _LEADERBOARD_SNAPSHOT_PREWARM_DAYS,
-        _LEADERBOARD_SNAPSHOT_BATCH_SIZE,
-        _LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS,
-    )
-    try:
-        if _LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS > 0:
-            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS)
-        while True:
-            try:
-                processed, queued = await _run_leaderboard_snapshot_tick()
-                if processed or queued:
-                    logger.info(
-                        "Leaderboard snapshot scheduler tick: queued=%d processed=%d",
-                        queued,
-                        processed,
-                    )
-            except Exception:
-                logger.exception("Leaderboard snapshot scheduler tick failed (non-fatal)")
-            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS)
-    finally:
-        _release_singleton_lock(lock_handle)
-
-
-def _warn_if_multi_worker() -> None:
+    finally:
+        _release_singleton_lock(lock_handle)
+
+
+def _env_bool(name: str, default: bool) -> bool:
+    raw = os.getenv(name)
+    if raw is None:
+        return default
+    return raw.strip().lower() not in {"0", "false", "no", "off"}
+
+
+def _env_int(name: str, default: int, *, alias: str | None = None, minimum: int = 1) -> int:
+    raw = os.getenv(name)
+    if raw is None and alias:
+        raw = os.getenv(alias)
+    if raw is None:
+        raw = str(default)
+    try:
+        value = int(raw)
+    except ValueError:
+        return default
+    return max(minimum, value)
+
+
+def _env_float(name: str, default: float, *, alias: str | None = None, minimum: float = 0.0) -> float:
+    raw = os.getenv(name)
+    if raw is None and alias:
+        raw = os.getenv(alias)
+    if raw is None:
+        raw = str(default)
+    try:
+        value = float(raw)
+    except ValueError:
+        return default
+    return max(minimum, value)
+
+
+_LEADERBOARD_SNAPSHOT_BACKGROUND_ENABLED = _env_bool("ADMIN_LEADERBOARD_BACKGROUND_ENABLED", True)
+_LEADERBOARD_SNAPSHOT_PREWARM_DAYS = _env_int(
+    "ADMIN_LEADERBOARD_BACKGROUND_PREWARM_DAYS",
+    30,
+)
+_LEADERBOARD_SNAPSHOT_BATCH_SIZE = _env_int(
+    "ADMIN_LEADERBOARD_BACKGROUND_BATCH_SIZE",
+    1,
+)
+_LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS = _env_float(
+    "ADMIN_LEADERBOARD_BACKGROUND_INTERVAL_SECONDS",
+    300.0,
+)
+_LEADERBOARD_SNAPSHOT_SLEEP_SECONDS = _env_float(
+    "ADMIN_LEADERBOARD_BACKGROUND_SLEEP_SECONDS",
+    2.0,
+)
+_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS = _env_int(
+    "ADMIN_LEADERBOARD_BACKGROUND_RUNNING_TTL_SECONDS",
+    1800,
+)
+_LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS = _env_float(
+    "ADMIN_LEADERBOARD_BACKGROUND_STARTUP_DELAY_SECONDS",
+    60.0,
+)
+
+
+def _recent_leaderboard_stat_dates(days: int) -> list[str]:
+    from datetime import datetime, timedelta
+
+    from deerflow.persistence.types import BEIJING_TZ
+
+    today = datetime.now(BEIJING_TZ).date()
+    start = today - timedelta(days=max(1, days) - 1)
+    return [(start + timedelta(days=i)).isoformat() for i in range((today - start).days + 1)]
+
+
+async def _leaderboard_snapshot_due_dates(
+    session_factory,
+    *,
+    settings_hash: str,
+    limit: int,
+    running_ttl_seconds: int,
+) -> list[str]:
+    from datetime import UTC, datetime, timedelta
+
+    from sqlalchemy import and_, or_, select
+
+    from deerflow.persistence.admin_stats.model import AdminLeaderboardDailyStatRow
+
+    cutoff = datetime.now(UTC) - timedelta(seconds=running_ttl_seconds)
+    async with session_factory() as session:
+        rows = (
+            await session.execute(
+                select(AdminLeaderboardDailyStatRow.stat_date)
+                .where(
+                    AdminLeaderboardDailyStatRow.settings_hash == settings_hash,
+                    or_(
+                        AdminLeaderboardDailyStatRow.status.in_(["queued", "error"]),
+                        and_(
+                            AdminLeaderboardDailyStatRow.status == "running",
+                            AdminLeaderboardDailyStatRow.updated_at < cutoff,
+                        ),
+                    ),
+                )
+                .order_by(AdminLeaderboardDailyStatRow.stat_date.desc())
+                .limit(limit)
+            )
+        ).all()
+    return [str(row[0]) for row in rows]
+
+
+async def _claim_leaderboard_snapshot(
+    session_factory,
+    *,
+    stat_date: str,
+    settings_hash: str,
+    running_ttl_seconds: int,
+) -> bool:
+    from datetime import UTC, datetime
+
+    from app.gateway.routers.admin_users import _snapshot_is_fresh
+    from deerflow.persistence.admin_stats.model import AdminLeaderboardDailyStatRow
+    from deerflow.persistence.types import BEIJING_TZ
+
+    async with session_factory() as session:
+        row = await session.get(
+            AdminLeaderboardDailyStatRow,
+            {"stat_date": stat_date, "settings_hash": settings_hash},
+        )
+        if row is None or _snapshot_is_fresh(row, stat_date):
+            return False
+        if row.status == "running":
+            updated_at = row.updated_at
+            if isinstance(updated_at, datetime):
+                if updated_at.tzinfo is None:
+                    updated_at = updated_at.replace(tzinfo=BEIJING_TZ)
+                age = (datetime.now(BEIJING_TZ) - updated_at.astimezone(BEIJING_TZ)).total_seconds()
+                if age < running_ttl_seconds:
+                    return False
+        row.status = "running"
+        row.message = "排行榜后台统计任务已认领。"
+        row.error = None
+        row.updated_at = datetime.now(UTC)
+        await session.commit()
+    return True
+
+
+async def _run_leaderboard_snapshot_tick() -> tuple[int, int]:
+    from app.gateway.routers.admin_users import (
+        _daily_job_key,
+        _queue_daily_leaderboard_snapshots,
+        _refresh_daily_leaderboard_snapshot,
+        _resolve_leaderboard_snapshot_scope,
+    )
+    from deerflow.config.system_settings import load_system_settings
+    from deerflow.persistence.engine import get_session_factory
+
+    session_factory = get_session_factory()
+    if session_factory is None:
+        return 0, 0
+
+    settings = load_system_settings().leaderboard
+    excluded, include_scheduled, include_admins, include_failed, settings_hash = _resolve_leaderboard_snapshot_scope(settings)
+    stat_dates = _recent_leaderboard_stat_dates(_LEADERBOARD_SNAPSHOT_PREWARM_DAYS)
+    queued_dates = await _queue_daily_leaderboard_snapshots(
+        session_factory,
+        stat_dates,
+        settings_hash,
+        message="排行榜后台统计任务定期预热/补算队列。",
+    )
+
+    due_dates = await _leaderboard_snapshot_due_dates(
+        session_factory,
+        settings_hash=settings_hash,
+        limit=_LEADERBOARD_SNAPSHOT_BATCH_SIZE,
+        running_ttl_seconds=_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS,
+    )
+    processed = 0
+    for stat_date in due_dates:
+        claimed = await _claim_leaderboard_snapshot(
+            session_factory,
+            stat_date=stat_date,
+            settings_hash=settings_hash,
+            running_ttl_seconds=_LEADERBOARD_SNAPSHOT_RUNNING_TTL_SECONDS,
+        )
+        if not claimed:
+            continue
+        logger.info("Leaderboard snapshot scheduler: building date=%s hash=%s", stat_date, settings_hash[:10])
+        await _refresh_daily_leaderboard_snapshot(
+            _daily_job_key(stat_date, settings_hash),
+            session_factory=session_factory,
+            settings=settings,
+            stat_date=stat_date,
+            settings_hash=settings_hash,
+            excluded=excluded,
+            include_scheduled=include_scheduled,
+            include_admins=include_admins,
+            include_failed=include_failed,
+        )
+        processed += 1
+        if _LEADERBOARD_SNAPSHOT_SLEEP_SECONDS > 0 and processed < len(due_dates):
+            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_SLEEP_SECONDS)
+    return processed, len(queued_dates)
+
+
+async def _leaderboard_snapshot_scheduler_loop() -> None:
+    """Background task inside the existing gateway process for daily snapshots."""
+    if not _LEADERBOARD_SNAPSHOT_BACKGROUND_ENABLED:
+        logger.info("Leaderboard snapshot scheduler disabled by ADMIN_LEADERBOARD_BACKGROUND_ENABLED")
+        return
+
+    from deerflow.config.runtime_paths import runtime_home
+
+    lock_handle = _acquire_singleton_lock(runtime_home() / ".leaderboard_snapshot_scheduler.lock")
+    if lock_handle is None:
+        logger.info("Leaderboard snapshot scheduler: another gateway worker holds the lock; this worker skips")
+        return
+
+    logger.info(
+        "Leaderboard snapshot scheduler started: prewarm_days=%d batch_size=%d interval=%.1fs",
+        _LEADERBOARD_SNAPSHOT_PREWARM_DAYS,
+        _LEADERBOARD_SNAPSHOT_BATCH_SIZE,
+        _LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS,
+    )
+    try:
+        if _LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS > 0:
+            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_STARTUP_DELAY_SECONDS)
+        while True:
+            try:
+                processed, queued = await _run_leaderboard_snapshot_tick()
+                if processed or queued:
+                    logger.info(
+                        "Leaderboard snapshot scheduler tick: queued=%d processed=%d",
+                        queued,
+                        processed,
+                    )
+            except Exception:
+                logger.exception("Leaderboard snapshot scheduler tick failed (non-fatal)")
+            await asyncio.sleep(_LEADERBOARD_SNAPSHOT_INTERVAL_SECONDS)
+    finally:
+        _release_singleton_lock(lock_handle)
+
+
+def _warn_if_multi_worker() -> None:
    """检测 uvicorn / gunicorn 多 worker 部署并打 ERROR 日志。

    主 agent / canvas / AI 写作流式响应在 Gateway 进程内通过 ``StreamBridge``
@@ -799,21 +801,21 @@ async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
        # 默认每天凌晨 3 点扫一次,删 retention_days(默认 7 天)以前的会话,联动
        # 清理 MySQL 业务表 + SQLite checkpointer 里的 thread 状态。多 worker
        # 下用文件锁选 leader,只有抢到锁的 worker 真正跑清理。
-        try:
-            from app.gateway.ai_writing_cleanup import cleanup_scheduler_loop
-
-            app.state.ai_writing_cleanup_task = asyncio.create_task(cleanup_scheduler_loop(app))
-        except Exception:
-            logger.exception("AI writing cleanup scheduler failed to start (non-fatal)")
-
-        try:
-            app.state.leaderboard_snapshot_scheduler_task = asyncio.create_task(_leaderboard_snapshot_scheduler_loop())
-        except Exception:
-            logger.exception("Leaderboard snapshot scheduler failed to start (non-fatal)")
-
-        # Start IM channel service if any channels are configured
-        try:
-            from app.channels.service import start_channel_service
+        try:
+            from app.gateway.ai_writing_cleanup import cleanup_scheduler_loop
+
+            app.state.ai_writing_cleanup_task = asyncio.create_task(cleanup_scheduler_loop(app))
+        except Exception:
+            logger.exception("AI writing cleanup scheduler failed to start (non-fatal)")
+
+        try:
+            app.state.leaderboard_snapshot_scheduler_task = asyncio.create_task(_leaderboard_snapshot_scheduler_loop())
+        except Exception:
+            logger.exception("Leaderboard snapshot scheduler failed to start (non-fatal)")
+
+        # Start IM channel service if any channels are configured
+        try:
+            from app.channels.service import start_channel_service

            channel_service = await start_channel_service(app.state.config)
            logger.info("Channel service started: %s", channel_service.get_status())
@@ -857,17 +859,17 @@ async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
            except asyncio.CancelledError:
                pass
            except Exception:
-                logger.exception("AI writing cleanup scheduler shutdown error")
-
-        leaderboard_snapshot_task = getattr(app.state, "leaderboard_snapshot_scheduler_task", None)
-        if leaderboard_snapshot_task is not None:
-            leaderboard_snapshot_task.cancel()
-            try:
-                await leaderboard_snapshot_task
-            except asyncio.CancelledError:
-                pass
-            except Exception:
-                logger.exception("Leaderboard snapshot scheduler shutdown error")
+                logger.exception("AI writing cleanup scheduler shutdown error")
+
+        leaderboard_snapshot_task = getattr(app.state, "leaderboard_snapshot_scheduler_task", None)
+        if leaderboard_snapshot_task is not None:
+            leaderboard_snapshot_task.cancel()
+            try:
+                await leaderboard_snapshot_task
+            except asyncio.CancelledError:
+                pass
+            except Exception:
+                logger.exception("Leaderboard snapshot scheduler shutdown error")

        # 阶段 7 删冗余：AI 写作 graph 不再自管 checkpointer / in-flight task；
        # thread 状态由 LangGraph Server 主 checkpointer 接管，lifespan 出口由
@@ -1122,6 +1124,10 @@ This gateway provides custom endpoints for models, MCP configuration, skills, an
    # Roundtable-planning business chains at /api/roundtable-chains
    app.include_router(roundtable_chains.router)

+    # Standalone parallel multi-agent panel: fan-out streaming + saved tasks
+    app.include_router(parallel_agents.router)
+    app.include_router(parallel_agent_tasks.router)
+
    # Roundtable-planning background jobs at /api/roundtable-jobs
    app.include_router(roundtable_jobs.router)

@@ -1159,10 +1165,10 @@ This gateway provides custom endpoints for models, MCP configuration, skills, an
    app.include_router(tool_metrics.router)

    # Channels API is mounted at /api/channels
-    app.include_router(channels.router)
-
-    # Admin checkpoint migration API: SQLite legacy checkpoints -> Postgres.
-    app.include_router(checkpoint_migration.router)
+    app.include_router(channels.router)
+
+    # Admin checkpoint migration API: SQLite legacy checkpoints -> Postgres.
+    app.include_router(checkpoint_migration.router)

    # Assistants compatibility API (LangGraph Platform stub)
    app.include_router(assistants_compat.router)
@@ -179,6 +179,10 @@ async def langgraph_runtime(app: FastAPI) -> AsyncGenerator[None, None]:

        app.state.roundtable_chain_store = make_roundtable_chain_store(sf)

+        from deerflow.persistence.parallel_agent_tasks import make_parallel_agent_task_store
+
+        app.state.parallel_agent_task_store = make_parallel_agent_task_store(sf)
+
        from deerflow.persistence.roundtable_jobs import (
            make_roundtable_job_store,
            register_default_store as _register_roundtable_job_default_store,
@@ -0,0 +1,148 @@
+"""CRUD API for parallel multi-agent task records (并行多智能体任务).
+
+A task = one saved run of the standalone parallel multi-agent panel: a task name
+ brief dispatched to several agents in parallel, plus the resulting message list.
+Bound to the requesting user (not the browser), so it follows the account across
+devices/sessions — same model as roundtable drafts.
+
+Routes (prefix ``/api/parallel-agent-tasks``):
+    GET    ""               list current user's tasks (lightweight meta, no messages)
+    POST   ""               create a task
+    GET    "/{task_id}"     one task (full, incl. messages)
+    PUT    "/{task_id}"     partial update (title / brief / agents / messages)
+    DELETE "/{task_id}"     delete
+"""
+
+from __future__ import annotations
+
+import logging
+from datetime import datetime
+from typing import Any
+from uuid import uuid4
+
+from fastapi import APIRouter, HTTPException, Request
+from pydantic import BaseModel, Field
+
+from deerflow.runtime.user_context import get_effective_user_id
+
+logger = logging.getLogger(__name__)
+router = APIRouter(prefix="/api/parallel-agent-tasks", tags=["parallel-agent-tasks"])
+
+
+# ── schemas ────────────────────────────────────────────────────────────────
+
+
+class TaskAgentSchema(BaseModel):
+    agent_id: str = Field(min_length=1, max_length=128)
+    name: str = ""
+    avatarType: str | None = None
+
+
+class TaskResponse(BaseModel):
+    id: str
+    title: str = ""
+    brief: str | None = None
+    agents: list[TaskAgentSchema] = []
+    # messages is an opaque dialogue list (mirrors the frontend Dialogue[] shape);
+    # absent on list responses (lightweight), present on get.
+    messages: list[dict[str, Any]] | None = None
+    created_at: datetime | str | None = None
+    updated_at: datetime | str | None = None
+
+
+class TaskListResponse(BaseModel):
+    tasks: list[TaskResponse]
+
+
+class TaskCreateRequest(BaseModel):
+    id: str | None = Field(default=None, max_length=64)
+    title: str = Field(default="", max_length=512)
+    brief: str | None = None
+    agents: list[TaskAgentSchema] = []
+    messages: list[dict[str, Any]] = []
+
+
+class TaskUpdateRequest(BaseModel):
+    # All optional → callers send only what changed.
+    title: str | None = Field(default=None, max_length=512)
+    brief: str | None = None
+    agents: list[TaskAgentSchema] | None = None
+    messages: list[dict[str, Any]] | None = None
+
+
+# ── helpers ──────────────────────────────────────────────────────────────
+
+
+def _current_user_id(request: Request) -> str:
+    user = getattr(request.state, "user", None)
+    if user is not None:
+        return str(user.id)
+    return get_effective_user_id()
+
+
+def _get_store(request: Request):
+    store = getattr(request.app.state, "parallel_agent_task_store", None)
+    if store is None:
+        raise HTTPException(status_code=503, detail="Parallel agent task store not available")
+    return store
+
+
+# ── routes ───────────────────────────────────────────────────────────────
+
+
+@router.get("", response_model=TaskListResponse)
+async def list_tasks(request: Request) -> TaskListResponse:
+    store = _get_store(request)
+    user_id = _current_user_id(request)
+    rows = await store.list_tasks(user_id)
+    return TaskListResponse(tasks=[TaskResponse(**r) for r in rows])
+
+
+@router.post("", response_model=TaskResponse, status_code=201)
+async def create_task(request: Request, body: TaskCreateRequest) -> TaskResponse:
+    store = _get_store(request)
+    user_id = _current_user_id(request)
+    row = await store.create_task(
+        user_id,
+        {
+            "id": (body.id or uuid4().hex),
+            "title": body.title.strip(),
+            "brief": body.brief,
+            "agents": [a.model_dump() for a in body.agents],
+            "messages": body.messages,
+        },
+    )
+    return TaskResponse(**row)
+
+
+@router.get("/{task_id}", response_model=TaskResponse)
+async def get_task(request: Request, task_id: str) -> TaskResponse:
+    store = _get_store(request)
+    user_id = _current_user_id(request)
+    row = await store.get_task(task_id, user_id)
+    if row is None:
+        raise HTTPException(status_code=404, detail="Task not found")
+    return TaskResponse(**row)
+
+
+@router.put("/{task_id}", response_model=TaskResponse)
+async def update_task(request: Request, task_id: str, body: TaskUpdateRequest) -> TaskResponse:
+    store = _get_store(request)
+    user_id = _current_user_id(request)
+    fields = body.model_dump(exclude_unset=True)
+    # Normalize agents (pydantic models) to plain dicts for JSON storage.
+    if "agents" in fields and fields["agents"] is not None:
+        fields["agents"] = [a if isinstance(a, dict) else a.model_dump() for a in body.agents or []]
+    row = await store.update_task(task_id, user_id, **fields)
+    if row is None:
+        raise HTTPException(status_code=404, detail="Task not found")
+    return TaskResponse(**row)
+
+
+@router.delete("/{task_id}", status_code=204)
+async def delete_task(request: Request, task_id: str) -> None:
+    store = _get_store(request)
+    user_id = _current_user_id(request)
+    deleted = await store.delete_task(task_id, user_id)
+    if not deleted:
+        raise HTTPException(status_code=404, detail="Task not found")
@@ -0,0 +1,247 @@
+"""Server-side parallel fan-out for the standalone parallel multi-agent panel.
+
+The Vue embeddable panel sends a single request — ``{task_name, brief, agent_ids}``
+— and this endpoint **fans out** to every agent in parallel (file writes disabled),
+then **merges** their LangGraph streams into ONE SSE tagged per ``agent_id``, so the
+client only consumes one stream and never parses LangGraph internals.
+
+It reuses the roundtable single-agent machinery from ``multi_agent``:
+``_special_run(..., no_file=True)`` runs one seat under the ``seat_parallel`` run
+policy (which excludes ``write_file`` / ``str_replace`` / ``bash`` / ``web_search``
+/ ``agent_orchestration`` …), so parallel agents can't write files or dispatch.
+
+Unified event protocol (one JSON object per ``data:`` frame)::
+
+    {"type": "agent_start", "agent_id": "..."}
+    {"type": "delta",       "agent_id": "...", "text": "..."}      # AI text increment
+    {"type": "step",        "agent_id": "...", "key": "...", "label": "...", "args": {...}}
+    {"type": "agent_done",  "agent_id": "...", "content": "..."}
+    {"type": "agent_error", "agent_id": "...", "message": "..."}
+    {"type": "all_done"}
+
+The frame-normalisation (LangGraph ``messages-tuple`` → delta / tool step) mirrors
+the frontend parser in ``frontend-web/src/roundtable-planning/api/multi-agent.ts``
+(``extractAiTextDelta`` + ``tool_call_chunks`` accumulation).
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+from typing import Any, AsyncIterator
+
+from fastapi import APIRouter, Body, HTTPException, Request
+from fastapi.responses import StreamingResponse
+
+from app.gateway.routers.multi_agent import (
+    _ROUNDTABLE_THREAD_METADATA,
+    _auth_headers,
+    _create_thread,
+    _loopback_base,
+    _make_loopback_client,
+    _special_run,
+    _validate_agent_id,
+)
+
+logger = logging.getLogger(__name__)
+router = APIRouter(prefix="/api/parallel-agents", tags=["parallel-agents"])
+
+# Soft cap on parallel degree — protects remote model concurrency + sandbox.
+MAX_PARALLEL_AGENTS = 8
+
+
+# ── frame normalisation (mirrors multi-agent.ts) ─────────────────────────────
+
+
+def _extract_ai_text(chunk: dict[str, Any]) -> str:
+    """AI text delta from a ``messages-tuple`` chunk obj; "" when not AI text."""
+    type_ = str(chunk.get("type") or "").lower()
+    if type_ not in ("ai", "aimessagechunk", "aimessage"):
+        return ""
+    content = chunk.get("content")
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        out: list[str] = []
+        for item in content:
+            if isinstance(item, str):
+                out.append(item)
+            elif isinstance(item, dict):
+                t = item.get("text")
+                if isinstance(t, str):
+                    out.append(t)
+        return "".join(out)
+    return ""
+
+
+def _extract_tool_steps(chunk: dict[str, Any]) -> list[dict[str, Any]]:
+    """Tool-call steps from a chunk obj — both complete ``tool_calls`` and
+    incremental ``tool_call_chunks`` (we only need the name to surface a step)."""
+    steps: list[dict[str, Any]] = []
+    tool_calls = chunk.get("tool_calls")
+    if isinstance(tool_calls, list):
+        for tc in tool_calls:
+            if not isinstance(tc, dict):
+                continue
+            name = tc.get("name")
+            if not isinstance(name, str) or not name:
+                continue
+            ident = tc.get("id") or name
+            args = tc.get("args") if isinstance(tc.get("args"), dict) else None
+            steps.append({"key": f"tool:{ident}", "label": f"调用工具 {name}", "args": args})
+    tcc = chunk.get("tool_call_chunks")
+    if isinstance(tcc, list):
+        for c in tcc:
+            if not isinstance(c, dict):
+                continue
+            name = c.get("name")
+            if not isinstance(name, str) or not name:
+                continue
+            idx = c.get("index") if isinstance(c.get("index"), int) else 0
+            ident = c.get("id") or f"{name}:{idx}"
+            steps.append({"key": f"tool:{ident}", "label": f"调用工具 {name}", "args": None})
+    return steps
+
+
+def _build_task_message(task_name: str, brief: str) -> str:
+    parts: list[str] = []
+    if task_name.strip():
+        parts.append(f"任务名称：{task_name.strip()}")
+    if brief.strip():
+        parts.append(f"任务简介：\n{brief.strip()}")
+    parts.append(
+        "请你从你的专业职责出发，针对以上任务给出你的分析与交付。"
+        "本轮为**并行**研讨：请把你的产出**直接写在回复正文里**，"
+        "不要写文件、不要执行脚本/命令。"
+    )
+    return "\n\n".join(parts)
+
+
+# ── endpoint ─────────────────────────────────────────────────────────────────
+
+
+@router.post("/run/stream")
+async def run_parallel_agents_stream(request: Request, payload: dict = Body(...)) -> StreamingResponse:
+    """并行扇出 N 个智能体，合并成一条带 ``agent_id`` 标记的 SSE。文件写入固定禁用。"""
+    task_name: str = str(payload.get("task_name") or "").strip()
+    brief: str = str(payload.get("brief") or "").strip()
+    raw_ids = payload.get("agent_ids") or []
+    model_name: str = str(payload.get("model") or "")
+
+    if not isinstance(raw_ids, list) or not raw_ids:
+        raise HTTPException(status_code=400, detail="agent_ids 不能为空")
+    agent_ids: list[str] = []
+    for aid in raw_ids:
+        sid = str(aid or "").strip().lower()
+        _validate_agent_id(sid, "agent_ids")
+        if sid not in agent_ids:
+            agent_ids.append(sid)
+    if len(agent_ids) > MAX_PARALLEL_AGENTS:
+        raise HTTPException(
+            status_code=400,
+            detail=f"并行智能体数量上限 {MAX_PARALLEL_AGENTS}（当前 {len(agent_ids)}）",
+        )
+    if not task_name and not brief:
+        raise HTTPException(status_code=400, detail="任务名称与简介至少填写其一")
+
+    base = _loopback_base(request)
+    headers = _auth_headers(request)
+    new_message = _build_task_message(task_name, brief)
+
+    async def _consume_agent(
+        client: Any,
+        agent_id: str,
+        queue: asyncio.Queue[dict[str, Any]],
+    ) -> None:
+        await queue.put({"type": "agent_start", "agent_id": agent_id})
+        last_event = ""
+        seen_steps: set[str] = set()
+        buffer: list[str] = []
+        done_sent = False
+        try:
+            thread_id = await _create_thread(client, base, headers, metadata=_ROUNDTABLE_THREAD_METADATA)
+            gen = _special_run(
+                client,
+                base,
+                headers,
+                {agent_id: thread_id},
+                agent_id,
+                new_message,
+                [],
+                model_name,
+                display_name="",
+                no_file=True,  # seat_parallel policy → 禁 write_file/bash/str_replace…
+            )
+            async for chunk in gen:
+                if isinstance(chunk, dict):
+                    # 终态帧 {status, content, agent_name}
+                    content = str(chunk.get("content") or "")
+                    await queue.put({"type": "agent_done", "agent_id": agent_id, "content": content})
+                    done_sent = True
+                    continue
+                line = chunk.strip() if isinstance(chunk, str) else ""
+                if line.startswith("event:"):
+                    last_event = line[6:].strip()
+                    continue
+                if not line.startswith("data:"):
+                    continue
+                raw = line[5:].strip()
+                if not raw:
+                    continue
+                try:
+                    parsed = json.loads(raw)
+                except Exception:
+                    continue
+                if last_event == "messages" and isinstance(parsed, list) and parsed:
+                    obj = parsed[0]
+                    if isinstance(obj, dict):
+                        text = _extract_ai_text(obj)
+                        if text:
+                            buffer.append(text)
+                            await queue.put({"type": "delta", "agent_id": agent_id, "text": text})
+                        for step in _extract_tool_steps(obj):
+                            key = step["key"]
+                            if key in seen_steps:
+                                continue
+                            seen_steps.add(key)
+                            await queue.put({"type": "step", "agent_id": agent_id, **step})
+            if not done_sent:
+                # 流被抢占/截断,没收到终态帧 —— 用已累积正文兜底收口本席位。
+                await queue.put(
+                    {"type": "agent_done", "agent_id": agent_id, "content": "".join(buffer)}
+                )
+        except Exception as exc:  # noqa: BLE001 — 单个 agent 失败不拖垮其它
+            logger.warning("[parallel-agents] %s failed: %r", agent_id, exc)
+            await queue.put({"type": "agent_error", "agent_id": agent_id, "message": str(exc)})
+        finally:
+            await queue.put({"type": "__agent_end__", "agent_id": agent_id})
+
+    async def stream_generator() -> AsyncIterator[str]:
+        queue: asyncio.Queue[dict[str, Any]] = asyncio.Queue()
+        async with _make_loopback_client(request) as client:
+            tasks = [asyncio.create_task(_consume_agent(client, aid, queue)) for aid in agent_ids]
+            ended = 0
+            try:
+                while ended < len(agent_ids):
+                    event = await queue.get()
+                    if event.get("type") == "__agent_end__":
+                        ended += 1
+                        continue
+                    yield f"data: {json.dumps(event, ensure_ascii=False)}\n\n"
+                yield f"data: {json.dumps({'type': 'all_done'}, ensure_ascii=False)}\n\n"
+            finally:
+                for t in tasks:
+                    if not t.done():
+                        t.cancel()
+                await asyncio.gather(*tasks, return_exceptions=True)
+
+    return StreamingResponse(
+        stream_generator(),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "X-Accel-Buffering": "no",
+            "Connection": "keep-alive",
+        },
+    )
@@ -0,0 +1,61 @@
+"""Create parallel_agent_tasks table.
+
+Saved runs of the standalone parallel multi-agent panel (Vue embeddable
+component): task name + brief + agents snapshot + the resulting message list,
+bound per user. ``brief`` / ``agents`` / ``messages`` are LONGTEXT on MySQL
+(mirror ``roundtable_drafts`` headroom) and plain TEXT elsewhere — matching
+``PortableLongText``. ``create_all`` builds the table for brand-new DBs; this
+revision patches DBs that already exist.
+
+Chained after ``20260608_04`` (chain stage_goals) so the revision graph stays
+single-headed — otherwise ``alembic upgrade head`` aborts with "multiple heads"
+and the startup auto-migrate silently falls back to the create_all schema.
+
+Revision ID: 20260608_05
+Revises: 20260608_04
+Create Date: 2026-06-08
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy import inspect
+from sqlalchemy.dialects.mysql import LONGTEXT
+
+
+revision = "20260608_05"
+down_revision = "20260608_04"
+branch_labels = None
+depends_on = None
+
+
+def _has_table(table_name: str) -> bool:
+    inspector = inspect(op.get_bind())
+    return table_name in inspector.get_table_names()
+
+
+def upgrade() -> None:
+    if _has_table("parallel_agent_tasks"):
+        return
+    long_text = LONGTEXT() if op.get_bind().dialect.name == "mysql" else sa.Text()
+    op.create_table(
+        "parallel_agent_tasks",
+        sa.Column("id", sa.String(length=64), primary_key=True, nullable=False),
+        sa.Column("user_id", sa.String(length=64), nullable=False),
+        sa.Column("title", sa.String(length=512), nullable=False, server_default=""),
+        sa.Column("brief", long_text, nullable=True),
+        sa.Column("agents", long_text, nullable=True),
+        sa.Column("messages", long_text, nullable=True),
+        sa.Column("created_at", sa.DateTime(), nullable=False),
+        sa.Column("updated_at", sa.DateTime(), nullable=False),
+    )
+    op.create_index(
+        "ix_parallel_agent_tasks_user_id", "parallel_agent_tasks", ["user_id"]
+    )
+
+
+def downgrade() -> None:
+    if _has_table("parallel_agent_tasks"):
+        op.drop_index("ix_parallel_agent_tasks_user_id", table_name="parallel_agent_tasks")
+        op.drop_table("parallel_agent_tasks")
@@ -27,6 +27,7 @@ from deerflow.persistence.feedback.model import FeedbackRow
 from deerflow.persistence.llm_metrics.model import LlmCallMetricRow
 from deerflow.persistence.models.run_event import RunEventRow
 from deerflow.persistence.notifications.model import NotificationRow
+from deerflow.persistence.parallel_agent_tasks.model import ParallelAgentTaskRow
 from deerflow.persistence.recommended_questions.model import RecommendedQuestionRow
 from deerflow.persistence.roundtable_chains.model import RoundtableChainRow
 from deerflow.persistence.roundtable_draft_shares.model import RoundtableDraftShareRow
@@ -62,6 +63,7 @@ __all__ = [
    "AdminLeaderboardDailyStatRow",
    "LlmCallMetricRow",
    "NotificationRow",
+    "ParallelAgentTaskRow",
    "RecommendedQuestionRow",
    "RoundtableChainRow",
    "RoundtableDraftRow",
@@ -0,0 +1,18 @@
+"""Parallel multi-agent task persistence."""
+
+from __future__ import annotations
+
+from deerflow.persistence.parallel_agent_tasks.model import ParallelAgentTaskRow
+from deerflow.persistence.parallel_agent_tasks.sql import ParallelAgentTaskRepository
+
+__all__ = [
+    "ParallelAgentTaskRepository",
+    "ParallelAgentTaskRow",
+    "make_parallel_agent_task_store",
+]
+
+
+def make_parallel_agent_task_store(session_factory):
+    if session_factory is None:
+        return None
+    return ParallelAgentTaskRepository(session_factory)
@@ -0,0 +1,46 @@
+"""ORM model for parallel multi-agent task records.
+
+A "parallel-agent task" is one saved run of the standalone *parallel multi-agent*
+panel (the Vue3 embeddable component): a task name + brief dispatched to several
+agents in parallel, plus the resulting chat/message list. Persisted per user so it
+follows the account across devices/sessions — same model as ``roundtable_drafts``.
+
+``agents`` and ``messages`` are stored as JSON serialized into ``PortableLongText``
+(MySQL ``LONGTEXT``) rather than ``PortableJSON`` (MySQL ``TEXT``, 64 KB cap),
+because the parallel transcript can carry several agents' full deliveries that
+easily exceed 64 KB — mirroring ``roundtable_drafts``.
+"""
+
+from __future__ import annotations
+
+from datetime import UTC, datetime
+
+from sqlalchemy import String
+from sqlalchemy.orm import Mapped, mapped_column
+
+from deerflow.persistence.base import Base
+from deerflow.persistence.types import BeijingDateTime, PortableLongText
+
+
+class ParallelAgentTaskRow(Base):
+    __tablename__ = "parallel_agent_tasks"
+
+    id: Mapped[str] = mapped_column(String(64), primary_key=True)
+    user_id: Mapped[str] = mapped_column(String(64), nullable=False, index=True)
+    # Task name (用户输入的任务名称) shown on the history list.
+    title: Mapped[str] = mapped_column(String(512), nullable=False, default="")
+    # Task brief / 简介 — the free-text task description.
+    brief: Mapped[str | None] = mapped_column(PortableLongText, nullable=True)
+    # JSON: snapshot of the agents dispatched, [{"agent_id", "name", "avatarType"?}].
+    agents: Mapped[str | None] = mapped_column(PortableLongText, nullable=True)
+    # JSON: the full message/dialogue list (mirrors the frontend Dialogue[] shape).
+    messages: Mapped[str | None] = mapped_column(PortableLongText, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(
+        BeijingDateTime(), nullable=False, default=lambda: datetime.now(UTC)
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        BeijingDateTime(),
+        nullable=False,
+        default=lambda: datetime.now(UTC),
+        onupdate=lambda: datetime.now(UTC),
+    )
@@ -0,0 +1,131 @@
+"""SQLAlchemy-backed store for parallel multi-agent task records.
+
+``agents`` / ``messages`` JSON columns are stored as serialized strings in
+``PortableLongText`` and round-tripped here with ``json.dumps`` / ``json.loads``
+— same approach as ``roundtable_drafts`` / ``roundtable_chains``.
+
+All write/read paths are scoped by ``user_id`` for ownership isolation; a row
+that belongs to another user is treated as not found.
+"""
+
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+from typing import Any
+
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
+
+from deerflow.persistence.parallel_agent_tasks.model import ParallelAgentTaskRow
+
+_UNSET = object()
+
+
+def _loads(value: Any) -> Any:
+    if not isinstance(value, str):
+        return value
+    if not value:
+        return None
+    try:
+        return json.loads(value)
+    except Exception:
+        return None
+
+
+def _dumps(value: Any) -> str | None:
+    if value is None:
+        return None
+    return json.dumps(value, ensure_ascii=False)
+
+
+class ParallelAgentTaskRepository:
+    def __init__(self, session_factory: async_sessionmaker[AsyncSession]) -> None:
+        self._sf = session_factory
+
+    @staticmethod
+    def _to_dict(row: ParallelAgentTaskRow, *, with_messages: bool = True) -> dict[str, Any]:
+        data: dict[str, Any] = {
+            "id": row.id,
+            "title": row.title,
+            "brief": row.brief,
+            "agents": _loads(row.agents) or [],
+            "created_at": row.created_at.isoformat() if isinstance(row.created_at, datetime) else row.created_at,
+            "updated_at": row.updated_at.isoformat() if isinstance(row.updated_at, datetime) else row.updated_at,
+        }
+        if with_messages:
+            data["messages"] = _loads(row.messages) or []
+        return data
+
+    async def list_tasks(self, user_id: str, *, limit: int = 200) -> list[dict[str, Any]]:
+        """列出当前用户的任务（轻量元信息，不含 messages，供历史列表）。"""
+        stmt = (
+            select(ParallelAgentTaskRow)
+            .where(ParallelAgentTaskRow.user_id == user_id)
+            .order_by(ParallelAgentTaskRow.updated_at.desc())
+            .limit(limit)
+        )
+        async with self._sf() as session:
+            result = await session.execute(stmt)
+            return [self._to_dict(row, with_messages=False) for row in result.scalars()]
+
+    async def get_task(self, task_id: str, user_id: str) -> dict[str, Any] | None:
+        async with self._sf() as session:
+            row = await session.get(ParallelAgentTaskRow, task_id)
+            if row is None or row.user_id != user_id:
+                return None
+            return self._to_dict(row)
+
+    async def create_task(self, user_id: str, data: dict[str, Any]) -> dict[str, Any]:
+        now = datetime.now(UTC)
+        row = ParallelAgentTaskRow(
+            id=data["id"],
+            user_id=user_id,
+            title=(data.get("title") or "").strip(),
+            brief=(data.get("brief") or None),
+            agents=_dumps(data.get("agents") or []),
+            messages=_dumps(data.get("messages") or []),
+            created_at=now,
+            updated_at=now,
+        )
+        async with self._sf() as session:
+            session.add(row)
+            await session.commit()
+            await session.refresh(row)
+            return self._to_dict(row)
+
+    async def update_task(
+        self,
+        task_id: str,
+        user_id: str,
+        *,
+        title=_UNSET,
+        brief=_UNSET,
+        agents=_UNSET,
+        messages=_UNSET,
+    ) -> dict[str, Any] | None:
+        async with self._sf() as session:
+            row = await session.get(ParallelAgentTaskRow, task_id)
+            if row is None or row.user_id != user_id:
+                return None
+            if title is not _UNSET and title is not None:
+                row.title = str(title).strip()
+            if brief is not _UNSET:
+                row.brief = brief or None
+            if agents is not _UNSET:
+                row.agents = _dumps(agents)
+            if messages is not _UNSET:
+                row.messages = _dumps(messages)
+            row.updated_at = datetime.now(UTC)
+            await session.commit()
+            await session.refresh(row)
+            return self._to_dict(row)
+
+    async def delete_task(self, task_id: str, user_id: str) -> bool:
+        async with self._sf() as session:
+            row = await session.get(ParallelAgentTaskRow, task_id)
+            if row is None or row.user_id != user_id:
+                return False
+            await session.delete(row)
+            await session.commit()
+            return True
@@ -0,0 +1,79 @@
+# parallel-agents-vue
+
+并行多智能体面板 —— Vue3 可嵌入组件。输入「任务名称 + 简介」，**并行调用多个智能体**
+（智能体清单由宿主传入，运行期间后端固定**禁止写文件**），消息列表照搬 DeerFlow 圆桌
+研讨第二步并行多席位的**深色样式与完整结构**（步骤条 / 并行双列 / 席位状态条），支持
+**保存聊天记录 + 回显**与**重新分析**。
+
+## 开发调试
+
+```bash
+pnpm install
+pnpm dev        # http://localhost:5183 ，填后端 baseUrl / token / 智能体 id 即可联调
+```
+
+后端需要一个跑着的 DeerFlow 网关（默认 `http://localhost:8001`），它提供：
+
+- `POST /api/parallel-agents/run/stream` —— 并行扇出，合并 SSE（事件带 `agent_id`）；
+- `/api/parallel-agent-tasks` —— 任务保存 / 回显 CRUD。
+
+## 构建为组件库
+
+```bash
+pnpm build:lib  # 产出 dist/parallel-agents-vue.{js,umd.cjs} + dist/parallel-agents-vue.css
+```
+
+## 嵌入其它 Vue 项目
+
+```ts
+import { ParallelAgentsPanel } from "parallel-agents-vue";
+import "parallel-agents-vue/style.css";
+```
+
+```vue
+<!-- 放进你自己的抽屉 / 弹窗里，面板会自适应容器（h-full w-full）。
+     接口地址走配置文件默认（src/config.ts），登录只需把用户名传进来（免密自动登录）。 -->
+<YourDrawer v-model="open">
+  <div style="height: 80vh">
+    <ParallelAgentsPanel
+      :username="currentUsername"
+      :agents="[
+        { agent_id: 'roundtable-intelligence', name: '情报收集' },
+        { agent_id: 'roundtable-risk-review', name: '风险审查' },
+      ]"
+      @saved="onSaved"
+    />
+  </div>
+</YourDrawer>
+```
+
+### 接口地址 & 登录
+
+- **接口地址**写在配置文件 `src/config.ts` 的 `DEFAULT_BASE_URL`（dev 可用 `VITE_PA_BASE_URL` 覆盖）；
+  也可用 `:base-url` prop 覆盖。
+- **登录**：把宿主侧的**用户名**通过 `:username` 传进来即可，组件会自动走 DeerFlow 的
+  **免密「用户名登录」**端点（`POST /api/v1/auth/login/username`，公共 + CSRF 豁免）换取 token
+  并缓存、自动续期（401 重登）。**不需要宿主自己管 token**。
+  - 若你已有 token，也可直接传 `:token` / `:get-token`（优先级高于 username）。
+  - 若后端开了口令门（`auth_login.username_login.require_password=true`），再传 `:login-password`。
+
+### Props
+
+| prop            | 类型                                   | 说明                                                  |
+| --------------- | -------------------------------------- | ----------------------------------------------------- |
+| `agents`        | `{agent_id,name,avatarType?}[]`（必填）| 要并行调用的智能体                                    |
+| `username`      | `string`                               | 免密自动登录用户名（推荐方式，宿主自己传）            |
+| `baseUrl`       | `string`                               | 后端地址；不传用 `config.ts` 默认                     |
+| `token`         | `string`                               | 已有 Bearer token（与 `username` 二选一，优先）       |
+| `getToken`      | `() => string \| Promise<string>`      | 动态取 token                                          |
+| `loginPassword` | `string`                               | 仅当后端开启口令门时需要                              |
+| `taskId`        | `string`                               | 传入则挂载时自动加载并回显该已保存任务                |
+| `model`         | `string`                               | 可选模型名（不传用 `config.ts` 默认 / 后端回退）      |
+| `dark`          | `boolean`（默认 `true`）               | 深色面板（自带成色，不依赖宿主主题）                  |
+
+### 事件
+
+- `@saved="(task) => ..."` —— 保存成功（含新建/更新）后回传 `SavedTask`。
+- `@update:taskId="(id) => ..."` —— 当前任务 id 变化（保存 / 加载历史时）。
+
+> 样式作用域全部包在根类 `.parallel-agents` 下，且**不引入 Tailwind preflight**，不会污染宿主全局样式。
@@ -0,0 +1,47 @@
+# 并行多智能体面板 · 测试用例总览
+
+本目录是 `parallel-agents-vue` 的**手工测试用例**（Markdown）。覆盖：并行调用与流式、文件写入禁用、
+保存与回显、重新分析、免密登录与鉴权、嵌入与深色样式。
+
+## 用例清单
+
+| 文件                                         | 主题                                  |
+| -------------------------------------------- | ------------------------------------- |
+| [01-并行调用与流式.md](./01-并行调用与流式.md) | 并行扇出、流式输出、步骤条、席位状态条、禁写文件 |
+| [02-保存与回显.md](./02-保存与回显.md)        | 保存聊天记录、历史列表、加载回显、更新保存 |
+| [03-重新分析.md](./03-重新分析.md)            | 重新分析（覆盖重跑）、停止、空输入校验 |
+| [04-免密登录与鉴权.md](./04-免密登录与鉴权.md) | username 免密登录、token 优先级、401 重登、口令门 |
+| [05-嵌入与深色样式.md](./05-嵌入与深色样式.md) | 配置文件接口地址、抽屉/弹窗嵌入、深色自适应、不污染宿主 |
+| [06-可用智能体清单.md](./06-可用智能体清单.md) | **测试用真实 agent_id 清单（可直接复制粘贴）** + 示例任务 |
+| [07-使用说明.md](./07-使用说明.md)            | **在其它 Vue 项目中引入哪些组件、如何使用**（props/事件/嵌入） |
+
+## 测试环境
+
+| 组件            | 地址 / 命令                                            |
+| --------------- | ------------------------------------------------------ |
+| DeerFlow 后端   | `http://localhost:8001`（`make dev`，需含本功能的新路由）|
+| Vue dev demo    | `http://localhost:5183`（`pnpm dev`）                  |
+| 接口地址配置    | `src/config.ts` 的 `DEFAULT_BASE_URL`                  |
+
+## 前置准备（所有用例通用）
+
+1. 后端已启动，且包含本功能：
+   - `POST /api/parallel-agents/run/stream`（并行扇出，合并 SSE）
+   - `GET/POST/PUT/DELETE /api/parallel-agent-tasks`（任务保存/回显）
+   - `POST /api/v1/auth/login/username`（免密登录，公共）
+   - 数据表 `parallel_agent_tasks` 已建（`alembic upgrade head` 或 `create_all`）。
+2. 至少准备 **2 个真实存在的 agent_id**（必须是后端注册过的真实 id，不能只写名字）。
+   本环境可直接用：`roundtable-solution-design`（方案设计）、`roundtable-risk-review`（风险审查）等
+   —— 完整可复制清单见 [06-可用智能体清单.md](./06-可用智能体清单.md)。
+3. 打开 dev demo `http://localhost:5183`，填：
+   - 后端 baseUrl：`http://localhost:8001`
+   - 用户名：`demo`（免密自动登录）
+   - 智能体：每行 `agent_id, 显示名`
+   点「打开面板（弹窗内）」。
+
+## 用例字段约定
+
+- **用例ID**：`PA-<模块>-<序号>`
+- **优先级**：P0（核心，必过）/ P1（重要）/ P2（边界）
+- **前置条件 → 操作步骤 → 预期结果**：逐条核对，全部命中才算通过。
+- **结果记录**：建议在「实际结果 / 通过?」列里回填。
@@ -0,0 +1,105 @@
+# 测试用例 · 并行调用与流式
+
+覆盖：并行扇出多个智能体、流式逐字输出、步骤条（工具调用）、点头像切换显示（单选）、席位状态条、
+运行期间**禁止写文件**。
+
+---
+
+## PA-RUN-01 · 并行调用多个智能体（P0）
+
+**前置条件**
+- 面板已打开，填入任务名「竞品调研」、简介「对比 A、B 两款产品的核心功能与定价」。
+- 传入 2 个智能体（情报收集、风险审查）。
+
+**操作步骤**
+1. 点击「开始分析」。
+
+**预期结果**
+- 顶部出现**席位状态条**，显示「并行 2 席」，两个席位头像角标先后转为「进行中」（转圈）。
+- 两个智能体**同时**开始执行（并行，而非一个跑完再跑下一个）——可在状态条上看到都在「进行中」。
+- 会话区每次显示**一个**席位的消息列表（默认第 1 个），点头像可切换查看另一个（见 PA-RUN-06）。
+- 气泡头部显示：头像、智能体名、状态徽标「并行执行中」、时间。
+
+---
+
+## PA-RUN-02 · 流式逐字输出 + 光标（P0）
+
+**前置条件**：PA-RUN-01 正在运行中。
+
+**操作步骤**
+1. 观察当前显示席位的气泡正文（点头像可切换看其它席位）。
+
+**预期结果**
+- 正文**逐步增量**出现（流式），不是一次性整段刷出。
+- 正在输出的气泡末尾有**闪烁光标**（`.pa-cursor`）。
+- 内容还没来时，显示 loading 占位（转圈 + 跳动圆点 + 「并行执行中」文案）。
+- Markdown 能正确渲染（标题/列表/代码块/表格按 `.pa-md` 样式）。
+
+---
+
+## PA-RUN-03 · 步骤条（思维链 / 工具调用）（P1）
+
+**前置条件**：智能体会调用工具（如检索类技能）。
+
+**操作步骤**
+1. 运行一轮，观察气泡顶部的「执行步骤」折叠卡。
+
+**预期结果**
+- 出现「执行步骤（N）」折叠卡（`MessageStepsCard`），点击可展开/收起。
+- 每条步骤显示：工具标签（如「调用工具 xxx」）、参数摘要（如有 `path` 优先显示）、时间。
+- 同一工具调用**不重复**出条目（按 key 去重）。
+- 若智能体本轮不调工具，则不出现步骤卡（属正常）。
+
+---
+
+## PA-RUN-04 · 席位状态流转（P1）
+
+**操作步骤**
+1. 运行一轮，全程观察席位状态条各头像角标。
+
+**预期结果**
+- 角标状态依次：灰（pending）→ 蓝色转圈（running）→ 绿色对勾（done）。
+- 某智能体失败时其角标变**红色叉**（error），且对应气泡显示「该智能体执行出错：…」、状态徽标「出错」。
+- 单个智能体失败**不影响**其它智能体继续完成（互不拖累）。
+- 全部结束后，运行按钮从「停止」恢复为「重新分析」。
+
+---
+
+## PA-RUN-05 · 运行期间禁止写文件（P0）
+
+**目的**：核对后端 `seat_parallel` 策略生效——并行席位被禁用 `write_file` / `str_replace` / `bash` 等。
+
+**操作步骤**
+1. 简介里**故意要求写文件**，例如：「请把结果保存成一个 .md 文件」。
+2. 点击「开始分析」，等待完成。
+
+**预期结果**
+- 智能体**不会**真正落地文件，产出**直接呈现在气泡正文**里。
+- 步骤条里**不出现** `write_file` / `bash` / `str_replace` 工具调用（被策略排除）。
+- 可旁证：后端日志中该席位 `role=seat_parallel`。
+
+---
+
+## PA-RUN-06 · 点头像切换显示（单选）（P1）
+
+**前置条件**：传入 ≥2 个智能体并运行。
+
+**操作步骤**
+1. 点击席位状态条上的某个头像。
+
+**预期结果**
+- 主区**切换为只显示**该智能体的消息列表（**单选**，每次显示一个）。
+- 被选中的头像高亮（外发光描边）；再点别的头像即切到那个。
+- 切换不影响后台并行执行：未显示的智能体仍在跑/已完成，其增量在切回时能看到。
+- 默认显示第 1 个席位。
+
+---
+
+## PA-RUN-07 · 后端不可达 / 智能体 id 非法（P2）
+
+**操作步骤**
+1. 把 baseUrl 改成一个打不通的地址，或传入不存在的 agent_id，点「开始分析」。
+
+**预期结果**
+- 顶部出现红色错误提示（请求失败信息），不白屏、不卡死。
+- 非法 agent_id 时，对应席位以 error 角标 + 出错气泡呈现，其它正常席位不受影响。
@@ -0,0 +1,94 @@
+# 测试用例 · 保存与回显
+
+覆盖：保存聊天记录（新建/更新）、历史列表、加载已存任务并回显、`taskId` 自动加载、跨刷新持久化。
+
+---
+
+## PA-SAVE-01 · 首次保存（新建）（P0）
+
+**前置条件**：已跑完一轮、会话区有结果。任务名「竞品调研」。
+
+**操作步骤**
+1. 点击「保存」。
+
+**预期结果**
+- 按钮短暂显示「保存中…」，成功后变为「更新保存」（说明已绑定 taskId）。
+- 触发 `@saved` 事件，回传 `SavedTask`（含 id、title、agents、messages）。
+- 「历史」下拉里出现这条记录（标题 + 更新时间）。
+- 后端 `parallel_agent_tasks` 表新增一行（owner = 当前登录用户）。
+
+---
+
+## PA-SAVE-02 · 再次保存（更新覆盖）（P0）
+
+**前置条件**：PA-SAVE-01 已保存、当前仍是同一任务。
+
+**操作步骤**
+1. 修改任务名 / 简介，或重新分析得到新结果。
+2. 点击「更新保存」。
+
+**预期结果**
+- **不新增**记录，而是更新原 taskId 那条（title / brief / agents / messages 被覆盖为最新）。
+- 历史列表里该条的更新时间刷新、排到最前。
+
+---
+
+## PA-SAVE-03 · 历史加载并回显（P0）
+
+**前置条件**：历史里至少有 1 条已保存任务。
+
+**操作步骤**
+1. 点击「历史 ▾」展开列表。
+2. 选择一条历史任务。
+
+**预期结果**
+- 任务名、简介**回填**为该任务保存时的值。
+- 会话区**完整回显**保存时的消息列表（各智能体气泡、正文、步骤条），且**均为非流式终态**（无光标、无 loading）。
+- 席位状态条据消息重建：正常席位显示 done（绿勾），出错的显示 error（红叉）。
+- 当前 taskId 切换为该任务（再点「保存」是「更新保存」该条）。
+
+---
+
+## PA-SAVE-04 · taskId 进入即自动回显（P1）
+
+**操作步骤**
+1. 以 `:task-id="<已存在任务id>"` 方式挂载面板（嵌入场景）。
+
+**预期结果**
+- 面板挂载后**自动加载**该任务并回显（等同于 PA-SAVE-03 的结果），无需手动点历史。
+- `@update:taskId` 随之触发。
+
+---
+
+## PA-SAVE-05 · 跨刷新持久化（P1）
+
+**操作步骤**
+1. 保存一条任务后，刷新浏览器（或重开宿主页面）。
+2. 用相同用户名进入，打开历史。
+
+**预期结果**
+- 历史里仍能看到之前保存的任务（数据存在后端、按用户隔离，非浏览器 localStorage）。
+- 加载后能正常回显。
+
+---
+
+## PA-SAVE-06 · 无结果时禁止保存（P2）
+
+**操作步骤**
+1. 新开面板、尚未运行任何分析。
+
+**预期结果**
+- 「保存」按钮**禁用**（无可保存内容）。
+- 运行中也不应因保存导致状态错乱（保存按钮在有结果后才可用）。
+
+---
+
+## PA-SAVE-07 · 用户隔离（P1）
+
+**操作步骤**
+1. 用户名 A 保存一条任务。
+2. 切换用户名为 B（重新进入面板），打开历史。
+
+**预期结果**
+- B 的历史里**看不到** A 的任务。
+- B 直接用 A 的 taskId 加载应失败（404 / 无内容），不报错崩溃。
@@ -0,0 +1,77 @@
+# 测试用例 · 重新分析与运行控制
+
+覆盖：重新分析（覆盖重跑）、运行中停止、空输入校验、运行态按钮切换。
+
+---
+
+## PA-RE-01 · 重新分析（覆盖重跑）（P0）
+
+**前置条件**：已跑完一轮、会话区有结果（按钮显示「重新分析」）。
+
+**操作步骤**
+1. （可选）修改任务名 / 简介。
+2. 点击「重新分析」。
+
+**预期结果**
+- **清空**上一轮的所有气泡与席位状态，按当前输入**重新并行**调用各智能体。
+- 不会和上一轮结果混在一起（旧消息被替换，不是追加）。
+- 重跑过程与首跑一致（流式、单列消息 + 点头像切换、步骤条、席位状态流转）。
+
+---
+
+## PA-RE-02 · 运行中按钮为「停止」（P1）
+
+**操作步骤**
+1. 开始分析，运行未结束时观察操作区。
+
+**预期结果**
+- 运行期间，主按钮显示「停止」（不再是「开始/重新分析」）。
+- 「保存」按钮在运行中可点（保存当前已产出的内容）或按实现禁用——以不报错为准。
+
+---
+
+## PA-RE-03 · 停止当前运行（P0）
+
+**前置条件**：正在运行中。
+
+**操作步骤**
+1. 点击「停止」。
+
+**预期结果**
+- 进行中的请求被取消（AbortController），不再有新增量进来。
+- 仍在 streaming 的气泡被收尾为终态（去掉光标，徽标置「已完成」）。
+- 按钮恢复为「重新分析」，可立即再次发起。
+- 控制台不应出现未捕获的 AbortError 报错（已被吞掉）。
+
+---
+
+## PA-RE-04 · 空输入校验（P1）
+
+**操作步骤**
+1. 任务名与简介**都留空**，点击「开始分析」。
+
+**预期结果**
+- 按钮处于**禁用**态，点击无效（`canStart` 为假需任务名或简介至少其一 + 至少一个智能体）。
+- 只填任务名或只填简介，按钮可用。
+
+---
+
+## PA-RE-05 · 未传智能体（P2）
+
+**操作步骤**
+1. `:agents="[]"`（空数组）挂载面板。
+
+**预期结果**
+- 操作区显示提示「⚠️ 未传入任何智能体（agents 为空）」。
+- 「开始分析」禁用，不会发起空请求。
+
+---
+
+## PA-RE-06 · 连续多次重新分析（P2）
+
+**操作步骤**
+1. 连续点击「重新分析」若干次（每次等上一轮结束或中途停止再点）。
+
+**预期结果**
+- 每次都干净重置 + 重跑，无残留旧气泡、无席位状态错乱。
+- 不出现多份 SSE 叠加导致的内容串列 / 重复。
@@ -0,0 +1,106 @@
+# 测试用例 · 免密登录与鉴权
+
+覆盖：username 免密自动登录、token 优先级、token 缓存与 401 重登、口令门、CSRF 绕过。
+
+> 背景：面板优先用宿主传入的 token；没有 token 时，用宿主传入的 **username** 走 DeerFlow 公共免密端点
+> `POST /api/v1/auth/login/username` 换取 `access_token`，以 `Authorization: Bearer` 发起后续请求
+> （带 Bearer 会绕过 CSRF）。token 解析优先级：`token` > `getToken` > 缓存 > `username` 登录。
+
+---
+
+## PA-AUTH-01 · username 免密自动登录（P0）
+
+**前置条件**：后端 `auth_login.username_login.require_password = false`（默认）。
+
+**操作步骤**
+1. 只传 `:username="demo"`（不传 token），打开面板。
+2. 开始分析 / 打开历史。
+
+**预期结果**
+- 面板能正常拉历史、能发起并行流式——说明已用 username 换到 token 并附带在请求头。
+- 旁证（手工 curl）：
+  ```bash
+  curl -s -X POST -H "Content-Type: application/json" \
+    -d '{"username":"demo"}' http://localhost:8001/api/v1/auth/login/username
+  # → 返回 {"access_token":"...","token_type":"bearer",...}
+  ```
+
+---
+
+## PA-AUTH-02 · token 优先级高于 username（P1）
+
+**操作步骤**
+1. 同时传 `:token="<有效token>"` 和 `:username="demo"`。
+2. 触发请求（开始分析 / 历史）。
+
+**预期结果**
+- 直接用传入的 `token`（不会再去调登录端点）。
+- `:get-token` 同理优先于 username（动态取 token）。
+
+---
+
+## PA-AUTH-03 · token 缓存（不重复登录）（P1）
+
+**操作步骤**
+1. 只传 username，先后触发多次请求（拉历史 → 开始分析 → 保存）。
+
+**预期结果**
+- 登录端点**只调用一次**（首个请求），后续请求复用缓存 token（按 `baseUrl::username` 缓存）。
+- 可在浏览器 Network 面板核对 `login/username` 只出现一次。
+
+---
+
+## PA-AUTH-04 · 401 自动重登重试（P1）
+
+**前置条件**：能制造 token 失效（例如等待过期，或后端清理）。
+
+**操作步骤**
+1. 用 username 登录并缓存 token 后，使其失效，再触发一次 CRUD 请求。
+
+**预期结果**
+- 首次请求收到 401 → 面板**清缓存并用 username 重登一次**，再自动重试该请求。
+- 用户无感知（不需手动重新登录），请求最终成功。
+
+---
+
+## PA-AUTH-05 · 口令门（require_password=true）（P2）
+
+**前置条件**：后端把 `auth_login.username_login.require_password` 设为 `true`、配置 `password`。
+
+**操作步骤**
+1. 只传 username、不传 `loginPassword` → 触发请求。
+2. 再传上正确的 `:login-password="<口令>"` → 触发请求。
+
+**预期结果**
+- 不传口令时：登录返回 401「口令错误」，面板顶部红色错误提示，不崩溃。
+- 传对口令时：登录成功，后续请求正常。
+
+---
+
+## PA-AUTH-06 · Bearer 绕过 CSRF（P1）
+
+**目的**：核对带 Bearer 的 POST 不被 CSRF 拦截。
+
+**操作步骤（手工 curl 旁证）**
+```bash
+TOKEN=...   # 上面登录拿到的 access_token
+# 带 Bearer 的 POST：参数非法应为 400/422，而非 403（403=CSRF 拦截）
+curl -s -o /dev/null -w "%{http_code}\n" -X POST \
+  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
+  -d '{}' http://localhost:8001/api/parallel-agents/run/stream
+```
+
+**预期结果**
+- 返回 **400/422**（业务参数校验），**不是 403**——证明 Bearer 已绕过 CSRF。
+- 不带任何鉴权时该 POST 返回 401/403（被拦截），属正常。
+
+---
+
+## PA-AUTH-07 · 未传 token 且未传 username（P2）
+
+**操作步骤**
+1. 既不传 token / getToken，也不传 username，触发需要鉴权的请求。
+
+**预期结果**
+- 请求因缺少鉴权失败（401/403），面板给出可读的错误提示（不白屏）。
+- 控制台/提示能反映「缺少 token 且未提供 username，无法登录」之类信息。
@@ -0,0 +1,105 @@
+# 测试用例 · 嵌入与深色样式
+
+覆盖：接口地址走配置文件、组件库构建产物、嵌入其它 Vue 项目、放进抽屉/弹窗自适应、深色成色、
+不污染宿主全局样式。
+
+---
+
+## PA-EMBED-01 · 接口地址走配置文件（P0）
+
+**操作步骤**
+1. 不传 `:base-url`，只改 `src/config.ts` 的 `DEFAULT_BASE_URL` 为目标后端地址。
+2. 运行面板触发请求。
+
+**预期结果**
+- 请求打到配置文件里的地址。
+- 传 `:base-url` prop 时**优先**用 prop（覆盖配置默认）。
+- dev 下设 `VITE_PA_BASE_URL` 环境变量也能覆盖默认。
+
+---
+
+## PA-EMBED-02 · 组件库构建产物（P0）
+
+**操作步骤**
+```bash
+pnpm build:lib
+```
+
+**预期结果**
+- `dist/` 产出：`parallel-agents-vue.js`（ESM）、`parallel-agents-vue.umd.cjs`（UMD）、`parallel-agents-vue.css`。
+- `vue` 被 external（不打进库，由宿主提供）。
+- 构建无报错、无「named + default exports」警告。
+
+---
+
+## PA-EMBED-03 · 嵌入其它 Vue 项目（P0）
+
+**操作步骤**
+1. 在另一个 Vue3 项目里：
+   ```ts
+   import { ParallelAgentsPanel } from "parallel-agents-vue";
+   import "parallel-agents-vue/style.css";
+   ```
+2. 模板中放置：
+   ```vue
+   <ParallelAgentsPanel :username="user" :agents="agents" />
+   ```
+
+**预期结果**
+- 组件正常渲染、可交互（深色面板，与 demo 一致）。
+- 只需 `:username` + `:agents` 两个 prop 即可跑起来（接口地址走配置）。
+- 也支持 `app.use(ParallelAgentsVue)` 全局注册后用 `<ParallelAgentsPanel/>`。
+
+---
+
+## PA-EMBED-04 · 放进抽屉 / 弹窗自适应（P0）
+
+**操作步骤**
+1. 把面板放进宿主的抽屉 / 对话框，外层给一个固定高度容器：
+   ```vue
+   <YourDrawer v-model="open">
+     <div style="height: 80vh"><ParallelAgentsPanel :username="user" :agents="agents" /></div>
+   </YourDrawer>
+   ```
+2. 改变容器尺寸 / 视口大小。
+
+**预期结果**
+- 面板**充满容器**（根元素 `h-full w-full flex flex-col min-h-0`），不溢出、不塌缩。
+- 会话区内部独立滚动，输入区与席位条固定不被挤掉。
+- 窄容器下消息列表正常显示、内部滚动，不溢出。
+
+---
+
+## PA-EMBED-05 · 深色成色自带、与宿主主题无关（P1）
+
+**操作步骤**
+1. 分别在**浅色**宿主页和**深色**宿主页里嵌入面板。
+
+**预期结果**
+- 面板始终是深蓝深色（自带 `--pa-*` token，作用域在 `.parallel-agents`），不随宿主明暗变化。
+- 文案、徽标、步骤卡、并行列底、滚动条等观感与圆桌 Step2 深色一致。
+- 传 `:dark="false"` 时按实现切换（如未实现浅色则保持深色，至少不报错）。
+
+---
+
+## PA-EMBED-06 · 不污染宿主全局样式（P0）
+
+**目的**：核对**未引入 Tailwind preflight**（不重置宿主全局标签样式）。
+
+**操作步骤**
+1. 在一个有自己样式的宿主页里嵌入面板（页面有 `h1/p/button/ul` 等默认样式）。
+
+**预期结果**
+- 嵌入后，宿主页面**自身**的 `h1/p/button/列表/边距` 等样式**不被改变**（没有被 Tailwind 的 reset 抹平）。
+- 面板内部样式正常（不依赖宿主提供 Tailwind）。
+- 类名都在 `.parallel-agents` 作用域下，不与宿主类名冲突。
+
+---
+
+## PA-EMBED-07 · 多实例并存（P2）
+
+**操作步骤**
+1. 同一页面放置 2 个面板实例（不同 agents / 不同 username）。
+
+**预期结果**
+- 两个实例各自独立运行、各自保存/回显，互不串流、互不串 token 缓存（缓存按 `baseUrl::username` 区分）。
@@ -0,0 +1,73 @@
+# 测试用 · 可用智能体清单（可直接复制粘贴）
+
+测试时面板需要传入**真实存在的 `agent_id`**（后端真正拿去跑的标识，不能只写名字）。
+下面是当前后端里**可直接用于测试**的内置智能体（已排除 intent / recommender / coordinator 这些系统角色）。
+
+> 注：`name` 只是显示用、可省略；`agent_id` 必须照抄。下列为本环境实测可用的内置智能体；
+> 你部署里的智能体可能不同，用文末方法自查。
+
+## 可用智能体
+
+| agent_id                    | 显示名   |
+| --------------------------- | -------- |
+| `roundtable-solution-design`| 方案设计 |
+| `roundtable-risk-review`    | 风险审查 |
+| `roundtable-execution-plan` | 执行规划 |
+| `roundtable-report`         | 方案总结 |
+| `demo-expert`               | 演示专家 |
+
+## A. 粘贴进 dev demo 页「智能体」输入框（每行 `agent_id, 显示名`）
+
+推荐至少 2 个以便点头像切换查看不同智能体：
+
+```
+roundtable-solution-design, 方案设计
+roundtable-risk-review, 风险审查
+```
+
+想多看几列就多贴几行：
+
+```
+roundtable-solution-design, 方案设计
+roundtable-risk-review, 风险审查
+roundtable-execution-plan, 执行规划
+```
+
+## B. 粘贴进组件 `:agents` prop（嵌入场景）
+
+```ts
+const agents = [
+  { agent_id: "roundtable-solution-design", name: "方案设计" },
+  { agent_id: "roundtable-risk-review", name: "风险审查" },
+  { agent_id: "roundtable-execution-plan", name: "执行规划" },
+];
+```
+
+```vue
+<ParallelAgentsPanel :username="'demo'" :agents="agents" />
+```
+
+## C. 配一个示例任务（直接粘贴到面板）
+
+- **任务名称**：`新功能上线方案评估`
+- **任务简介**：
+  ```
+  我们要给 App 上线「扫码点餐」功能。请从你的专业角度，评估方案可行性、列出关键风险、
+  并给出落地步骤与里程碑。
+  ```
+
+## 自查：列出你部署里所有可用的 agent_id
+
+```bash
+# 1) 免密登录拿 token（用户名换成你的）
+TOKEN=$(curl -s -X POST -H "Content-Type: application/json" \
+  -d '{"username":"demo"}' http://localhost:8001/api/v1/auth/login/username \
+  | python -c "import sys,json;print(json.load(sys.stdin)['access_token'])")
+
+# 2) 列出 agent_id <TAB> name
+curl -s -H "Authorization: Bearer $TOKEN" http://localhost:8001/api/agents -o agents.json
+python -c "import json,io;[print(a['id'],'\t',a['name']) for a in json.load(io.open('agents.json',encoding='utf-8')).get('agents',[])]"
+```
+
+第一列 `id` 才是要填的 `agent_id`。挑你需要的智能体即可（避开 `roundtable-intent` /
+`roundtable-recommender` / `roundtable-coordinator*` 这些系统角色——它们不是用来当并行席位的）。
@@ -0,0 +1,163 @@
+# 使用说明 · 在其它 Vue 项目中引入
+
+本组件库只对外暴露**一个**组件 `ParallelAgentsPanel`（其余 `MessageBubble` / `MessageStepsCard` /
+`ParallelSeatStrip` / `Avatar` / `MarkdownContent` 都是它的内部子组件，**无需也不要单独引入**）。
+宿主只管把它放进自己的页面 / 抽屉 / 弹窗，并传几个 prop。
+
+---
+
+## 一、要引入的东西（就这两样）
+
+1. **组件**：`ParallelAgentsPanel`
+2. **样式**：`parallel-agents-vue/style.css`（一次性引入即可，样式作用域全在 `.parallel-agents` 下）
+
+> 依赖说明：`vue ^3.4`（宿主提供，库不打包 vue）；`markdown-it` 已打进库内，无需宿主安装。
+
+---
+
+## 二、两种引入方式
+
+### 方式 A：构建成库后按包引入（推荐）
+
+1. 在本项目构建：
+   ```bash
+   pnpm install
+   pnpm build:lib      # 产出 dist/parallel-agents-vue.{js,umd.cjs,css}
+   ```
+2. 宿主项目里安装（任选其一）：
+   - 把本目录作为本地依赖：`"parallel-agents-vue": "file:../parallel-agents-vue"`；
+   - 或发到你们的私有 npm 后正常 `pnpm add parallel-agents-vue`。
+3. 代码里引入：
+   ```ts
+   import { ParallelAgentsPanel } from "parallel-agents-vue";
+   import "parallel-agents-vue/style.css";
+   ```
+
+### 方式 B：直接拷贝源码（无需发包）
+
+把本项目 `src/` 下这些拷进宿主项目（保持相对结构），按相对路径引入 `ParallelAgentsPanel.vue` +
+`styles/parallel-agents.css` 即可：
+
+```
+src/
+  components/   ParallelAgentsPanel.vue · MessageBubble.vue · MessageStepsCard.vue
+                ParallelSeatStrip.vue · Avatar.vue · MarkdownContent.vue
+  composables/  useParallelRun.ts
+  lib/          avatar-constants.ts
+  styles/       parallel-agents.css
+  api.ts · config.ts · types.ts
+```
+宿主需有 `markdown-it` 依赖（`pnpm add markdown-it`），并保证构建支持 `.vue`（Vite/@vitejs/plugin-vue）。
+
+---
+
+## 三、注册与使用
+
+### 局部引入（推荐）
+
+```vue
+<script setup lang="ts">
+import { ParallelAgentsPanel } from "parallel-agents-vue";
+import "parallel-agents-vue/style.css";
+
+const agents = [
+  { agent_id: "roundtable-solution-design", name: "方案设计" },
+  { agent_id: "roundtable-risk-review", name: "风险审查" },
+];
+</script>
+
+<template>
+  <div style="height: 80vh">
+    <ParallelAgentsPanel :username="'demo'" :agents="agents" />
+  </div>
+</template>
+```
+
+### 全局注册（可选）
+
+```ts
+import ParallelAgentsVue from "parallel-agents-vue";
+import "parallel-agents-vue/style.css";
+app.use(ParallelAgentsVue); // 之后任意模板可直接用 <ParallelAgentsPanel/>
+```
+
+---
+
+## 四、放进抽屉 / 弹窗
+
+组件根元素是 `h-full w-full flex flex-col min-h-0`，会**充满父容器**。所以**父容器要给一个高度**
+（弹窗/抽屉本身有高度即可）：
+
+```vue
+<YourDrawer v-model="open" size="large">
+  <div style="height: 86vh">
+    <ParallelAgentsPanel :username="currentUsername" :agents="agents" @saved="onSaved" />
+  </div>
+</YourDrawer>
+```
+
+---
+
+## 五、接口地址与登录
+
+- **接口地址**：默认读 `src/config.ts` 的 `DEFAULT_BASE_URL`（可改源码，或 dev 设 `VITE_PA_BASE_URL`）；
+  也可用 `:base-url` prop 临时覆盖。
+- **登录**：把宿主的**用户名**用 `:username` 传进来即可——组件自动走 DeerFlow 免密登录端点换 token、
+  缓存、401 自动重登。**宿主不用管 token**。若已有 token，可改传 `:token` / `:get-token`（优先级更高）。
+- token 缓存按 `baseUrl::username` 维度，多个实例/多用户互不串。
+
+---
+
+## 六、Props
+
+| prop            | 类型                                    | 必填 | 说明                                            |
+| --------------- | --------------------------------------- | :--: | ----------------------------------------------- |
+| `agents`        | `{ agent_id, name, avatarType? }[]`     |  ✔   | 要并行调用的智能体（`agent_id` 必须是真实 id）  |
+| `username`      | `string`                                |      | 免密自动登录用户名（推荐方式）                  |
+| `baseUrl`       | `string`                                |      | 后端地址；不传用 `config.ts` 默认               |
+| `token`         | `string`                                |      | 已有 Bearer token（优先于 username）            |
+| `getToken`      | `() => string \| Promise<string>`       |      | 动态取 token                                    |
+| `loginPassword` | `string`                                |      | 仅当后端开启口令门时需要                        |
+| `taskId`        | `string`                                |      | 传入则挂载时自动加载并回显该任务                |
+| `model`         | `string`                                |      | 可选模型名（不传走默认 / 后端回退）             |
+| `dark`          | `boolean`（默认 `true`）                |      | 深色面板（自带成色，不依赖宿主主题）            |
+
+`AgentInput`/`SavedTask` 等类型也从包里导出：`import type { AgentInput, SavedTask } from "parallel-agents-vue"`。
+
+## 七、事件
+
+| 事件              | 载荷         | 触发时机                                |
+| ----------------- | ------------ | --------------------------------------- |
+| `saved`           | `SavedTask`  | 保存成功（新建或更新）后                |
+| `update:taskId`   | `string`     | 当前任务 id 变化（保存 / 加载历史时）   |
+
+```vue
+<ParallelAgentsPanel
+  :username="user"
+  :agents="agents"
+  :task-id="taskId"
+  @saved="(t) => console.log('saved', t.id)"
+  @update:taskId="(id) => (taskId = id)"
+/>
+```
+
+---
+
+## 八、交互要点（与圆桌 Step2 一致）
+
+- 输入「任务名称 + 简介」→「开始分析」→ 各智能体**并行**输出，运行期间**禁止写文件**（产出走正文）。
+- 主区每次只显示**一个**席位的消息列表；**点席位头像切换**要看哪个（单选）。
+- 顶部席位状态条显示每个智能体的进行中 / 完成 / 出错。
+- 操作区右侧有**模型切换**下拉（拉取后端 `/api/models`）；选「默认」则用 `model` prop / 配置 / 后端回退。
+  切换作用于**下一次**「开始分析 / 重新分析」，运行中禁用。
+- 「保存」存档；「历史」里选回**回显**；「重新分析」清空重跑。
+
+---
+
+## 九、注意事项
+
+- **不会污染宿主全局样式**：库未引入 Tailwind preflight，类名都在 `.parallel-agents` 作用域内。
+- **深色自带成色**：面板始终深蓝深色，不随宿主明暗切换（与圆桌 Step2 观感一致）。
+- **跨域**：若宿主与后端不同源，后端需允许该来源的 CORS + 带 `credentials`；登录端点公共且 CSRF 豁免，
+  业务请求用 Bearer 头已绕过 CSRF。
+- **agent_id 必须真实**：见 [06-可用智能体清单.md](./06-可用智能体清单.md) 获取可用 id 与自查方法。
@@ -0,0 +1,12 @@
+<!doctype html>
+<html lang="zh-CN">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>并行多智能体面板 · Dev</title>
+  </head>
+  <body style="margin: 0">
+    <div id="app"></div>
+    <script type="module" src="/src/main.ts"></script>
+  </body>
+</html>
@@ -0,0 +1,41 @@
+{
+  "name": "parallel-agents-vue",
+  "version": "0.1.0",
+  "description": "并行多智能体面板 —— Vue3 可嵌入组件（输入任务名+简介，并行调用多个智能体，深色消息列表 + 步骤条 + 保存回显 + 重新分析）",
+  "type": "module",
+  "main": "./dist/parallel-agents-vue.umd.cjs",
+  "module": "./dist/parallel-agents-vue.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/parallel-agents-vue.js",
+      "require": "./dist/parallel-agents-vue.umd.cjs"
+    },
+    "./style.css": "./dist/parallel-agents-vue.css"
+  },
+  "files": ["dist"],
+  "scripts": {
+    "dev": "vite",
+    "build": "vue-tsc --noEmit && vite build",
+    "build:lib": "vite build --mode lib",
+    "typecheck": "vue-tsc --noEmit",
+    "preview": "vite preview"
+  },
+  "peerDependencies": {
+    "vue": "^3.4.0"
+  },
+  "dependencies": {
+    "markdown-it": "^14.1.0"
+  },
+  "devDependencies": {
+    "@tailwindcss/vite": "^4.1.0",
+    "@types/markdown-it": "^14.1.2",
+    "@vitejs/plugin-vue": "^5.2.0",
+    "tailwindcss": "^4.1.0",
+    "typescript": "^5.6.0",
+    "vite": "^6.0.0",
+    "vue": "^3.5.0",
+    "vue-tsc": "^2.1.10"
+  }
+}
@@ -0,0 +1,72 @@
+<script setup lang="ts">
+/** Dev demo 调试页：填后端地址 / token / 智能体 id，把面板塞进一个弹窗，验证嵌入与流式。 */
+import { reactive, ref } from "vue";
+import ParallelAgentsPanel from "./components/ParallelAgentsPanel.vue";
+import type { AgentInput } from "./types";
+
+import { DEFAULT_BASE_URL } from "./config";
+
+const cfg = reactive({
+  baseUrl: localStorage.getItem("pa.baseUrl") || DEFAULT_BASE_URL,
+  username: localStorage.getItem("pa.username") || "demo",
+  agentsRaw: localStorage.getItem("pa.agents") || "",
+});
+
+const open = ref(false);
+const agents = ref<AgentInput[]>([]);
+
+function parseAgents(): AgentInput[] {
+  // 每行一个：`agent_id, 显示名` 或仅 `agent_id`。
+  return cfg.agentsRaw
+    .split("\n")
+    .map((l) => l.trim())
+    .filter(Boolean)
+    .map((l) => {
+      const [id, name] = l.split(",").map((x) => x.trim());
+      return { agent_id: id, name: name || id };
+    });
+}
+
+function launch() {
+  localStorage.setItem("pa.baseUrl", cfg.baseUrl);
+  localStorage.setItem("pa.username", cfg.username);
+  localStorage.setItem("pa.agents", cfg.agentsRaw);
+  agents.value = parseAgents();
+  open.value = true;
+}
+</script>
+
+<template>
+  <div style="min-height: 100vh; background: #0a1020; color: #cbd5e1; padding: 24px; font-family: system-ui">
+    <h1 style="font-size: 18px; font-weight: 700; margin-bottom: 12px">并行多智能体面板 · Dev Demo</h1>
+    <div style="display: flex; flex-direction: column; gap: 8px; max-width: 640px">
+      <label>后端 baseUrl
+        <input v-model="cfg.baseUrl" style="width: 100%; padding: 6px; background: #111c33; color: #e5edf7; border: 1px solid #2d4473; border-radius: 6px" />
+      </label>
+      <label>用户名（免密自动登录，由你自己传）
+        <input v-model="cfg.username" style="width: 100%; padding: 6px; background: #111c33; color: #e5edf7; border: 1px solid #2d4473; border-radius: 6px" />
+      </label>
+      <label>智能体（每行一个：agent_id, 显示名）
+        <textarea v-model="cfg.agentsRaw" rows="4" placeholder="roundtable-intelligence, 情报收集&#10;roundtable-risk-review, 风险审查" style="width: 100%; padding: 6px; background: #111c33; color: #e5edf7; border: 1px solid #2d4473; border-radius: 6px" />
+      </label>
+      <button @click="launch" style="align-self: flex-start; padding: 8px 16px; background: #1677ff; color: #fff; border: 0; border-radius: 6px; cursor: pointer">
+        打开面板（弹窗内）
+      </button>
+    </div>
+
+    <!-- 模拟宿主弹窗：面板自适应容器 -->
+    <div
+      v-if="open"
+      style="position: fixed; inset: 0; background: rgba(0,0,0,0.5); display: flex; align-items: center; justify-content: center; z-index: 50"
+      @click.self="open = false"
+    >
+      <div style="width: 92vw; max-width: 1080px; height: 86vh; border-radius: 12px; overflow: hidden; box-shadow: 0 20px 60px rgba(0,0,0,0.5)">
+        <ParallelAgentsPanel
+          :base-url="cfg.baseUrl"
+          :username="cfg.username || undefined"
+          :agents="agents"
+        />
+      </div>
+    </div>
+  </div>
+</template>
@@ -0,0 +1,236 @@
+/**
+ * 后端访问层：鉴权（宿主传入 token）+ 并行扇出 SSE + 任务 CRUD。
+ *
+ * 全部打到 DeerFlow 网关（宿主传入 `baseUrl`，如 http://localhost:8001）：
+ *  - `POST {baseUrl}/api/parallel-agents/run/stream` —— 并行扇出，合并 SSE；
+ *  - `{baseUrl}/api/parallel-agent-tasks` —— 任务保存/回显 CRUD。
+ */
+import type { AgentInput, Dialogue, SavedTask, StreamEvent } from "./types";
+import { USERNAME_LOGIN_PATH } from "./config";
+
+export interface ApiConfig {
+  baseUrl: string;
+  /** 已有登录态 token（Bearer）。可传字符串或一个（异步）取 token 的函数。 */
+  token?: string;
+  getToken?: () => string | Promise<string>;
+  /**
+   * 免密自动登录用户名：不传 token 时，组件用它走「用户名登录」（公共端点）换取 token
+   * 并缓存。用户名由宿主自己传。
+   */
+  username?: string;
+  /** 仅当后端 `auth_login.username_login.require_password=true` 时需要的口令。 */
+  loginPassword?: string;
+}
+
+function base(cfg: ApiConfig): string {
+  return cfg.baseUrl.replace(/\/$/, "");
+}
+
+// 免密登录拿到的 token 按 baseUrl::username 缓存，避免每个请求都登一次。
+const _tokenCache = new Map<string, string>();
+const cacheKey = (cfg: ApiConfig) => `${base(cfg)}::${cfg.username ?? ""}`;
+
+/** 用 username 走公共「用户名登录」端点换 access_token（CSRF 豁免）。 */
+async function loginByUsername(cfg: ApiConfig): Promise<string> {
+  if (!cfg.username) throw new Error("缺少 token 且未提供 username，无法登录");
+  let url = `${base(cfg)}${USERNAME_LOGIN_PATH}`;
+  if (cfg.loginPassword) url += `?password=${encodeURIComponent(cfg.loginPassword)}`;
+  const res = await fetch(url, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({ username: cfg.username }),
+    credentials: "include",
+  });
+  if (!res.ok) {
+    const text = await res.text().catch(() => "");
+    throw new Error(`username login failed (${res.status}): ${text}`);
+  }
+  const data = (await res.json()) as { access_token: string; token_type?: string };
+  const token = data.access_token;
+  _tokenCache.set(cacheKey(cfg), token);
+  return token;
+}
+
+/** token 解析优先级：显式 token > getToken > 缓存 > 用 username 免密登录。 */
+async function resolveToken(cfg: ApiConfig): Promise<string | undefined> {
+  if (cfg.token) return cfg.token;
+  if (cfg.getToken) return await cfg.getToken();
+  const cached = _tokenCache.get(cacheKey(cfg));
+  if (cached) return cached;
+  if (cfg.username) return await loginByUsername(cfg);
+  return undefined;
+}
+
+/** 清掉缓存 token（401 时强制下次重登）。 */
+function invalidateToken(cfg: ApiConfig): void {
+  _tokenCache.delete(cacheKey(cfg));
+}
+
+async function authHeaders(cfg: ApiConfig, json = true): Promise<Record<string, string>> {
+  const h: Record<string, string> = {};
+  if (json) h["Content-Type"] = "application/json";
+  const token = await resolveToken(cfg);
+  if (token) h["Authorization"] = token.startsWith("Bearer ") ? token : `Bearer ${token}`;
+  return h;
+}
+
+/** 带鉴权的 fetch；若 401 且用 username 免密登录，则清缓存重登并重试一次。 */
+async function authedFetch(
+  cfg: ApiConfig,
+  url: string,
+  init: RequestInit & { json?: boolean } = {},
+): Promise<Response> {
+  const { json = true, headers, ...rest } = init;
+  const run = async () =>
+    fetch(url, {
+      ...rest,
+      headers: { ...(await authHeaders(cfg, json)), ...(headers as Record<string, string> | undefined) },
+      credentials: "include",
+    });
+  let res = await run();
+  if (res.status === 401 && !cfg.token && !cfg.getToken && cfg.username) {
+    invalidateToken(cfg);
+    res = await run();
+  }
+  return res;
+}
+
+// ── 并行扇出 SSE ──────────────────────────────────────────────────────────────
+
+export interface StreamHandlers {
+  onEvent: (ev: StreamEvent) => void;
+  onError?: (err: unknown) => void;
+}
+
+/**
+ * 打开 `/api/parallel-agents/run/stream` 一条合并 SSE，逐条把统一事件回调出去。
+ * 返回的 Promise 在流结束（all_done / 连接关闭）后 resolve。用 `signal` 取消。
+ */
+export async function streamParallelAgents(
+  cfg: ApiConfig,
+  body: { task_name: string; brief: string; agent_ids: string[]; model?: string },
+  handlers: StreamHandlers,
+  signal?: AbortSignal,
+): Promise<void> {
+  const res = await authedFetch(cfg, `${base(cfg)}/api/parallel-agents/run/stream`, {
+    method: "POST",
+    body: JSON.stringify(body),
+    signal,
+  });
+  if (!res.ok || !res.body) {
+    const text = await res.text().catch(() => "");
+    throw new Error(`run/stream failed (${res.status}): ${text}`);
+  }
+  const reader = res.body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+  try {
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split("\n");
+      buffer = lines.pop() ?? "";
+      for (const raw of lines) {
+        const line = raw.trim();
+        if (!line.startsWith("data:")) continue;
+        const payload = line.slice(5).trim();
+        if (!payload) continue;
+        try {
+          handlers.onEvent(JSON.parse(payload) as StreamEvent);
+        } catch {
+          /* 忽略坏帧 */
+        }
+      }
+    }
+  } catch (err) {
+    if ((err as { name?: string })?.name !== "AbortError") handlers.onError?.(err);
+  } finally {
+    reader.releaseLock();
+  }
+}
+
+// ── 任务 CRUD（保存 / 回显）─────────────────────────────────────────────────
+
+interface ServerTask {
+  id: string;
+  title: string;
+  brief: string | null;
+  agents: AgentInput[];
+  messages?: Dialogue[];
+  created_at: string | null;
+  updated_at: string | null;
+}
+
+const TASKS = (cfg: ApiConfig) => `${base(cfg)}/api/parallel-agent-tasks`;
+
+async function readJson<T>(res: Response, ctx: string): Promise<T> {
+  if (!res.ok) {
+    const text = await res.text().catch(() => "");
+    throw new Error(`${ctx} failed (${res.status}): ${text}`);
+  }
+  return (await res.json()) as T;
+}
+
+export async function listTasks(cfg: ApiConfig): Promise<SavedTask[]> {
+  const res = await authedFetch(cfg, TASKS(cfg), { json: false });
+  const data = await readJson<{ tasks: ServerTask[] }>(res, "list tasks");
+  return data.tasks ?? [];
+}
+
+export async function getTask(cfg: ApiConfig, id: string): Promise<SavedTask | null> {
+  const res = await authedFetch(cfg, `${TASKS(cfg)}/${encodeURIComponent(id)}`, { json: false });
+  if (res.status === 404) return null;
+  return readJson<SavedTask>(res, "get task");
+}
+
+export async function createTask(
+  cfg: ApiConfig,
+  payload: { id?: string; title: string; brief?: string; agents: AgentInput[]; messages: Dialogue[] },
+): Promise<SavedTask> {
+  const res = await authedFetch(cfg, TASKS(cfg), { method: "POST", body: JSON.stringify(payload) });
+  return readJson<SavedTask>(res, "create task");
+}
+
+export async function updateTask(
+  cfg: ApiConfig,
+  id: string,
+  patch: { title?: string; brief?: string; agents?: AgentInput[]; messages?: Dialogue[] },
+): Promise<SavedTask | null> {
+  const res = await authedFetch(cfg, `${TASKS(cfg)}/${encodeURIComponent(id)}`, {
+    method: "PUT",
+    body: JSON.stringify(patch),
+  });
+  if (res.status === 404) return null;
+  return readJson<SavedTask>(res, "update task");
+}
+
+// ── 模型列表 ──────────────────────────────────────────────────────────────
+
+export interface ModelItem {
+  name: string;
+  display_name?: string;
+}
+
+/** 列出后端可用模型（GET /api/models）。失败返回空数组（不阻断主流程）。 */
+export async function listModels(cfg: ApiConfig): Promise<ModelItem[]> {
+  try {
+    const res = await authedFetch(cfg, `${base(cfg)}/api/models`, { json: false });
+    if (!res.ok) return [];
+    const data = (await res.json()) as { models?: ModelItem[] };
+    return data.models ?? [];
+  } catch {
+    return [];
+  }
+}
+
+export async function deleteTask(cfg: ApiConfig, id: string): Promise<void> {
+  const res = await authedFetch(cfg, `${TASKS(cfg)}/${encodeURIComponent(id)}`, {
+    method: "DELETE",
+    json: false,
+  });
+  if (!res.ok && res.status !== 404) {
+    const text = await res.text().catch(() => "");
+    throw new Error(`delete task failed (${res.status}): ${text}`);
+  }
+}
@@ -0,0 +1,50 @@
+<script setup lang="ts">
+import { computed } from "vue";
+import { avatarColor, isHuman } from "@/lib/avatar-constants";
+
+const props = withDefaults(defineProps<{ type: string; size?: number }>(), { size: 36 });
+const color = computed(() => avatarColor(props.type));
+const human = computed(() => isHuman(props.type));
+</script>
+
+<template>
+  <span
+    class="inline-flex shrink-0 items-center justify-center rounded-xl"
+    :style="{ width: `${size}px`, height: `${size}px`, backgroundColor: color.bg }"
+  >
+    <!-- 真人 -->
+    <svg
+      v-if="human"
+      :width="size * 0.58"
+      :height="size * 0.58"
+      viewBox="0 0 24 24"
+      fill="none"
+      :stroke="color.fg"
+      stroke-width="2"
+      stroke-linecap="round"
+      stroke-linejoin="round"
+    >
+      <circle cx="12" cy="8" r="4" />
+      <path d="M4 21c0-4 4-6 8-6s8 2 8 6" />
+    </svg>
+    <!-- 机器人 -->
+    <svg
+      v-else
+      :width="size * 0.6"
+      :height="size * 0.6"
+      viewBox="0 0 24 24"
+      fill="none"
+      :stroke="color.fg"
+      stroke-width="2"
+      stroke-linecap="round"
+      stroke-linejoin="round"
+    >
+      <rect x="4" y="8" width="16" height="11" rx="3" />
+      <path d="M12 8V4" />
+      <circle cx="12" cy="3" r="1" :fill="color.fg" />
+      <circle cx="9" cy="13" r="1.2" :fill="color.fg" stroke="none" />
+      <circle cx="15" cy="13" r="1.2" :fill="color.fg" stroke="none" />
+      <path d="M9 16.5h6" />
+    </svg>
+  </span>
+</template>
@@ -0,0 +1,13 @@
+<script setup lang="ts">
+import { computed } from "vue";
+import MarkdownIt from "markdown-it";
+
+const props = defineProps<{ source: string }>();
+
+const md = new MarkdownIt({ html: false, linkify: true, breaks: true });
+const html = computed(() => md.render(props.source ?? ""));
+</script>
+
+<template>
+  <div class="pa-md" v-html="html" />
+</template>
@@ -0,0 +1,50 @@
+<script setup lang="ts">
+/** 单条消息气泡（对齐 Step2 MessageBubble + Step2Message 结构）：头像 / 发言人 / 状态徽标 /
+ *  时间 / 步骤条 / Markdown 正文 / 流式光标 / loading 占位。 */
+import { computed } from "vue";
+import type { Dialogue } from "@/types";
+import Avatar from "./Avatar.vue";
+import MessageStepsCard from "./MessageStepsCard.vue";
+import MarkdownContent from "./MarkdownContent.vue";
+
+const props = defineProps<{ msg: Dialogue }>();
+const hasContent = computed(() => (props.msg.content ?? "").trim().length > 0);
+</script>
+
+<template>
+  <div class="flex items-start gap-3">
+    <Avatar :type="msg.avatarType" :size="36" />
+    <div class="min-w-0 flex-1 space-y-1">
+      <div class="flex items-center justify-between gap-2">
+        <div class="flex items-center gap-2">
+          <span class="text-sm font-extrabold pa-text">{{ msg.sender }}</span>
+          <span v-if="msg.confidence" class="pa-badge">{{ msg.confidence }}</span>
+        </div>
+        <span class="shrink-0 font-mono text-[11px] pa-text-muted">{{ msg.time }}</span>
+      </div>
+
+      <MessageStepsCard
+        v-if="msg.steps && msg.steps.length"
+        :steps="msg.steps"
+        :streaming="msg.streaming"
+      />
+
+      <div class="pa-text break-words">
+        <div
+          v-if="msg.streaming && !hasContent"
+          class="flex items-center gap-2 py-1 text-sm pa-text-muted"
+        >
+          <span class="pa-spin inline-block">◜</span>
+          <span>{{ msg.confidence || "正在生成…" }}</span>
+          <span class="inline-flex items-center gap-0.5">
+            <span class="pa-bounce inline-block size-1 rounded-full" style="background: var(--pa-primary)" />
+            <span class="pa-bounce inline-block size-1 rounded-full" style="background: var(--pa-primary); animation-delay: 0.15s" />
+            <span class="pa-bounce inline-block size-1 rounded-full" style="background: var(--pa-primary); animation-delay: 0.3s" />
+          </span>
+        </div>
+        <MarkdownContent v-if="hasContent" :source="msg.content" />
+        <span v-if="msg.streaming && hasContent" class="pa-cursor" />
+      </div>
+    </div>
+  </div>
+</template>
@@ -0,0 +1,52 @@
+<script setup lang="ts">
+/** 步骤条 / 思维链卡：折叠展示该气泡运行期间的工具调用步骤（对齐 Step2 MessageStepsCard）。 */
+import { ref } from "vue";
+import type { DialogueStep } from "@/types";
+
+defineProps<{ steps: DialogueStep[]; streaming?: boolean }>();
+const open = ref(true);
+
+function argSummary(args?: Record<string, unknown> | null): string {
+  if (!args) return "";
+  const path = (args as { path?: unknown }).path;
+  if (typeof path === "string") return path;
+  try {
+    const s = JSON.stringify(args);
+    return s.length > 60 ? s.slice(0, 60) + "…" : s;
+  } catch {
+    return "";
+  }
+}
+</script>
+
+<template>
+  <div class="pa-surface pa-border mb-2 rounded-lg border p-0.5">
+    <button
+      type="button"
+      class="flex w-full items-center gap-1.5 rounded-md px-2 py-1 text-left text-[11px] font-semibold pa-text-muted"
+      @click="open = !open"
+    >
+      <span
+        class="inline-block transition-transform"
+        :class="open ? 'rotate-90' : ''"
+        >▶</span
+      >
+      <span>执行步骤（{{ steps.length }}）</span>
+      <span v-if="streaming" class="pa-spin inline-block">◜</span>
+    </button>
+    <ol v-show="open" class="space-y-0.5 px-2 pb-1.5 pt-0.5">
+      <li
+        v-for="s in steps"
+        :key="s.id"
+        class="flex items-start gap-2 text-[11px] pa-text"
+      >
+        <span class="mt-1 inline-block size-1.5 shrink-0 rounded-full" style="background: var(--pa-primary)" />
+        <span class="min-w-0">
+          <span class="font-semibold">{{ s.label }}</span>
+          <span v-if="argSummary(s.args)" class="pa-text-muted"> · {{ argSummary(s.args) }}</span>
+          <span class="pa-text-muted"> · {{ s.ts }}</span>
+        </span>
+      </li>
+    </ol>
+  </div>
+</template>
@@ -0,0 +1,329 @@
+<script setup lang="ts">
+/**
+ * 并行多智能体面板（主组件，可嵌入）。
+ *
+ * 输入「任务名称 + 简介」，并行调用宿主传入的多个智能体（后端固定禁写文件），消息列表
+ * 照搬圆桌 Step2 并行多席位的深色样式与结构（步骤条 / 并行双列 / 席位状态条），支持保存
+ * 聊天记录 + 回显 + 重新分析。
+ *
+ * 自适应容器：根元素 `h-full w-full flex flex-col min-h-0`，宿主直接塞进抽屉 / 弹窗即可。
+ */
+import { computed, onMounted, reactive, ref, watch } from "vue";
+import type { AgentInput, Dialogue, SavedTask } from "@/types";
+import { useParallelRun } from "@/composables/useParallelRun";
+import { createTask, getTask, listModels, listTasks, updateTask, type ApiConfig, type ModelItem } from "@/api";
+import { DEFAULT_BASE_URL, DEFAULT_MODEL } from "@/config";
+import ParallelSeatStrip from "./ParallelSeatStrip.vue";
+import MessageBubble from "./MessageBubble.vue";
+
+const props = withDefaults(
+  defineProps<{
+    /** 后端地址；不传用配置文件默认（config.ts 的 DEFAULT_BASE_URL）。 */
+    baseUrl?: string;
+    /** 已有 token（Bearer），与 username 二选一；都不传走配置默认地址 + 免密登录失败提示。 */
+    token?: string;
+    getToken?: () => string | Promise<string>;
+    /** 免密自动登录用户名（由宿主自己传）。不传 token 时用它登录换 token。 */
+    username?: string;
+    /** 仅当后端开启口令门时需要。 */
+    loginPassword?: string;
+    agents: AgentInput[];
+    taskId?: string;
+    model?: string;
+    dark?: boolean;
+  }>(),
+  { dark: true },
+);
+
+const emit = defineEmits<{
+  (e: "saved", task: SavedTask): void;
+  (e: "update:taskId", id: string): void;
+}>();
+
+const cfg = computed<ApiConfig>(() => ({
+  baseUrl: props.baseUrl || DEFAULT_BASE_URL,
+  token: props.token,
+  getToken: props.getToken,
+  username: props.username,
+  loginPassword: props.loginPassword,
+}));
+
+const run = useParallelRun();
+const { dialogues, seats, running } = run;
+
+const form = reactive({ taskName: "", brief: "" });
+const currentTaskId = ref<string | null>(props.taskId ?? null);
+const selected = ref<string[]>([]);
+const saving = ref(false);
+const history = ref<SavedTask[]>([]);
+const showHistory = ref(false);
+const errorMsg = ref<string | null>(null);
+
+// 模型切换：拉后端模型列表，空字符串 = 用默认（prop/config/后端回退）。
+const models = ref<ModelItem[]>([]);
+const selectedModel = ref<string>(props.model ?? DEFAULT_MODEL ?? "");
+
+const dialogueById = computed(() => {
+  const m = new Map<number, Dialogue>();
+  for (const d of dialogues.value) m.set(d.id, d);
+  return m;
+});
+
+const hasResults = computed(() => dialogues.value.length > 0);
+const canStart = computed(
+  () => !running.value && props.agents.length > 0 && (form.taskName.trim() || form.brief.trim()),
+);
+
+// 单选：主区每次只显示**一个**席位的消息列表，点头像切换。
+watch(
+  seats,
+  (s) => {
+    if (!s.length) {
+      selected.value = [];
+      return;
+    }
+    // 保留仍存在的当前选中；首次 / 失效时默认显示第 1 个席位。
+    const cur = selected.value[0];
+    const keep = cur && s.some((x) => x.agentId === cur) ? cur : s[0].agentId;
+    selected.value = [keep];
+  },
+  { immediate: true },
+);
+
+/** 点头像切换显示：单选，始终切到点击的那个席位。 */
+function toggleSeat(agentId: string) {
+  selected.value = [agentId];
+}
+
+const visibleSeats = computed(() => {
+  const id = selected.value[0] ?? seats.value[0]?.agentId;
+  const seat = seats.value.find((s) => s.agentId === id);
+  return seat ? [seat] : [];
+});
+
+async function onStart() {
+  if (!canStart.value) return;
+  errorMsg.value = null;
+  try {
+    await run.start(cfg.value, {
+      taskName: form.taskName,
+      brief: form.brief,
+      agents: props.agents,
+      model: selectedModel.value || undefined,
+    });
+  } catch (err) {
+    errorMsg.value = err instanceof Error ? err.message : String(err);
+  }
+}
+
+async function onSave() {
+  if (saving.value || !hasResults.value) return;
+  saving.value = true;
+  errorMsg.value = null;
+  try {
+    const title = form.taskName.trim() || "未命名任务";
+    const agentsSnapshot: AgentInput[] = props.agents.map((a) => ({ ...a }));
+    let saved: SavedTask | null;
+    if (currentTaskId.value) {
+      saved = await updateTask(cfg.value, currentTaskId.value, {
+        title,
+        brief: form.brief,
+        agents: agentsSnapshot,
+        messages: dialogues.value,
+      });
+    } else {
+      saved = await createTask(cfg.value, {
+        title,
+        brief: form.brief,
+        agents: agentsSnapshot,
+        messages: dialogues.value,
+      });
+    }
+    if (saved) {
+      currentTaskId.value = saved.id;
+      emit("update:taskId", saved.id);
+      emit("saved", saved);
+      await refreshHistory();
+    }
+  } catch (err) {
+    errorMsg.value = err instanceof Error ? err.message : String(err);
+  } finally {
+    saving.value = false;
+  }
+}
+
+async function refreshHistory() {
+  try {
+    history.value = await listTasks(cfg.value);
+  } catch {
+    /* 历史拉取失败不阻断主流程 */
+  }
+}
+
+async function loadTask(id: string) {
+  showHistory.value = false;
+  errorMsg.value = null;
+  try {
+    const task = await getTask(cfg.value, id);
+    if (!task) return;
+    currentTaskId.value = task.id;
+    form.taskName = task.title ?? "";
+    form.brief = task.brief ?? "";
+    selected.value = [];
+    run.hydrate(task.messages ?? []);
+    emit("update:taskId", task.id);
+  } catch (err) {
+    errorMsg.value = err instanceof Error ? err.message : String(err);
+  }
+}
+
+async function refreshModels() {
+  const list = await listModels(cfg.value);
+  models.value = list;
+  // 当前选择不在列表里（且非空）时，回退为默认（空 = 后端回退）。
+  if (selectedModel.value && !list.some((m) => m.name === selectedModel.value)) {
+    selectedModel.value = "";
+  }
+}
+
+onMounted(() => {
+  void refreshHistory();
+  void refreshModels();
+  if (props.taskId) void loadTask(props.taskId);
+});
+</script>
+
+<template>
+  <div
+    class="parallel-agents pa-bg flex h-full w-full min-h-0 flex-col"
+    :class="dark ? 'parallel-agents--dark' : ''"
+  >
+    <!-- 顶部：任务名 + 操作 -->
+    <div class="pa-border flex shrink-0 flex-col gap-2 border-b px-4 py-3">
+      <div class="flex items-center gap-2">
+        <input
+          v-model="form.taskName"
+          class="pa-input min-w-0 flex-1 px-3 py-1.5 text-sm"
+          placeholder="任务名称"
+        />
+        <div class="relative shrink-0">
+          <button
+            type="button"
+            class="pa-btn-ghost px-2.5 py-1.5 text-xs"
+            @click="showHistory = !showHistory"
+          >
+            历史 ▾
+          </button>
+          <div
+            v-if="showHistory"
+            class="pa-card pa-border absolute right-0 z-20 mt-1 max-h-72 w-72 overflow-auto rounded-lg border p-1 shadow-lg"
+          >
+            <div v-if="!history.length" class="px-3 py-4 text-center text-xs pa-text-muted">
+              暂无保存的任务
+            </div>
+            <button
+              v-for="t in history"
+              :key="t.id"
+              type="button"
+              class="pa-btn-ghost block w-full truncate px-2.5 py-1.5 text-left text-xs"
+              style="border: 0"
+              @click="loadTask(t.id)"
+            >
+              <span class="font-semibold">{{ t.title || "未命名任务" }}</span>
+              <span class="pa-text-muted"> · {{ (t.updated_at || "").slice(0, 16).replace("T", " ") }}</span>
+            </button>
+          </div>
+        </div>
+      </div>
+      <textarea
+        v-model="form.brief"
+        rows="2"
+        class="pa-input w-full resize-y px-3 py-1.5 text-sm"
+        placeholder="任务简介：把背景、目标、约束写清楚，会并行下发给各智能体"
+      />
+      <div class="flex flex-wrap items-center gap-2">
+        <button
+          v-if="!running"
+          type="button"
+          class="pa-btn-primary px-3.5 py-1.5 text-sm font-semibold"
+          :disabled="!canStart"
+          @click="onStart"
+        >
+          {{ hasResults ? "重新分析" : "开始分析" }}
+        </button>
+        <button
+          v-else
+          type="button"
+          class="pa-btn-ghost px-3.5 py-1.5 text-sm font-semibold"
+          @click="run.cancel()"
+        >
+          停止
+        </button>
+        <button
+          type="button"
+          class="pa-btn-ghost px-3.5 py-1.5 text-sm font-semibold"
+          :disabled="saving || !hasResults"
+          @click="onSave"
+        >
+          {{ saving ? "保存中…" : currentTaskId ? "更新保存" : "保存" }}
+        </button>
+
+        <!-- 模型切换：作用于下一次「开始分析 / 重新分析」 -->
+        <label class="ml-auto flex items-center gap-1.5 text-xs pa-text-muted">
+          模型
+          <select
+            v-model="selectedModel"
+            :disabled="running"
+            class="pa-input px-2 py-1 text-xs"
+            title="选择本次并行分析使用的模型"
+          >
+            <option value="">默认</option>
+            <option v-for="m in models" :key="m.name" :value="m.name">
+              {{ m.display_name || m.name }}
+            </option>
+          </select>
+        </label>
+
+        <span v-if="props.agents.length === 0" class="text-xs pa-text-muted">
+          ⚠️ 未传入任何智能体（agents 为空）
+        </span>
+      </div>
+      <p v-if="errorMsg" class="text-xs" style="color: var(--pa-rose)">{{ errorMsg }}</p>
+    </div>
+
+    <!-- 席位状态条 -->
+    <div v-if="seats.length" class="pa-border shrink-0 border-b px-4">
+      <ParallelSeatStrip :seats="seats" :selected="selected" @toggle="toggleSeat" />
+    </div>
+
+    <!-- 会话区：并行双列 -->
+    <div class="min-h-0 flex-1 overflow-y-auto p-4">
+      <div
+        v-if="hasResults"
+        class="grid gap-4"
+        :class="visibleSeats.length > 1 ? 'grid-cols-1 lg:grid-cols-2' : 'grid-cols-1'"
+      >
+        <div
+          v-for="seat in visibleSeats"
+          :key="seat.agentId"
+          class="pa-surface-soft pa-border flex min-w-0 flex-col rounded-xl border p-3"
+        >
+          <MessageBubble
+            v-if="dialogueById.get(seat.bubbleId)"
+            :msg="dialogueById.get(seat.bubbleId)!"
+          />
+          <div v-else class="py-6 text-center text-xs pa-text-muted">
+            {{ seat.name }} · 等待输出…
+          </div>
+        </div>
+      </div>
+      <div
+        v-else
+        class="flex h-full flex-col items-center justify-center gap-2 text-center"
+      >
+        <p class="text-sm pa-text-muted">填写任务名称与简介，点「开始分析」并行调用各智能体。</p>
+        <p class="text-xs pa-text-muted">并行运行期间禁止写文件，产出直接呈现在正文。</p>
+      </div>
+    </div>
+  </div>
+</template>
@@ -0,0 +1,60 @@
+<script setup lang="ts">
+/** 并行席位状态条（对齐 Step2 ParallelSeatStrip）：席位头像 + running/done/error 角标，
+ *  点击可切换主区显示哪两列。 */
+import { computed } from "vue";
+import type { SeatView } from "@/types";
+import Avatar from "./Avatar.vue";
+
+const props = defineProps<{ seats: SeatView[]; selected: string[] }>();
+const emit = defineEmits<{ (e: "toggle", agentId: string): void }>();
+
+const runningCount = computed(() => props.seats.filter((s) => s.status === "running").length);
+</script>
+
+<template>
+  <div class="flex items-center gap-2 overflow-x-auto py-1">
+    <span class="shrink-0 text-[11px] font-bold pa-text-muted">
+      并行 {{ seats.length }} 席
+      <span v-if="runningCount > 0" class="pa-text-primary">（{{ runningCount }} 进行中）</span>
+      <span v-if="seats.length > 1" class="font-normal pa-text-muted">· 点头像切换显示</span>
+    </span>
+    <div class="flex items-center gap-2">
+      <button
+        v-for="seat in seats"
+        :key="seat.agentId"
+        type="button"
+        class="relative shrink-0 rounded-full transition"
+        :style="
+          selected.includes(seat.agentId)
+            ? 'box-shadow: 0 0 0 2px var(--pa-primary)'
+            : ''
+        "
+        :title="seat.name"
+        @click="emit('toggle', seat.agentId)"
+      >
+        <span
+          v-if="seat.status === 'running'"
+          class="pa-ping absolute inset-0 rounded-full"
+          style="background: rgba(22, 119, 255, 0.3)"
+        />
+        <span class="relative block" :style="!selected.includes(seat.agentId) ? 'opacity:0.8' : ''">
+          <Avatar :type="seat.avatarType" :size="28" />
+        </span>
+        <span
+          class="absolute -bottom-0.5 -right-0.5 flex size-3.5 items-center justify-center rounded-full text-[8px] font-black"
+          style="border: 1px solid var(--pa-bg)"
+          :class="{
+            'pa-dot-running': seat.status === 'running',
+            'pa-dot-done': seat.status === 'done',
+            'pa-dot-error': seat.status === 'error',
+          }"
+          :style="seat.status === 'pending' ? 'background: var(--pa-border)' : ''"
+        >
+          <span v-if="seat.status === 'running'" class="pa-spin">◜</span>
+          <span v-else-if="seat.status === 'done'">✓</span>
+          <span v-else-if="seat.status === 'error'">✕</span>
+        </span>
+      </button>
+    </div>
+  </div>
+</template>
@@ -0,0 +1,212 @@
+/**
+ * 并行运行编排（前端态）。消费后端合并 SSE，按 agent_id 把增量/步骤/收口路由到对应席位
+ * 气泡，维护并行席位状态条；带轻量节流（rAF 合并增量），支持取消、回显 hydrate、重置。
+ */
+import { ref, shallowRef } from "vue";
+import type { AgentInput, Dialogue, SeatView, StreamEvent } from "@/types";
+import { getAvatarTypeFor } from "@/lib/avatar-constants";
+import { streamParallelAgents, type ApiConfig } from "@/api";
+
+function nowTime(): string {
+  const d = new Date();
+  const p = (n: number) => String(n).padStart(2, "0");
+  return `${p(d.getHours())}:${p(d.getMinutes())}:${p(d.getSeconds())}`;
+}
+
+let _seq = 1;
+function nextId(): number {
+  return _seq++;
+}
+
+export function useParallelRun() {
+  const dialogues = ref<Dialogue[]>([]);
+  const seats = ref<SeatView[]>([]);
+  const running = ref(false);
+  const abortRef = shallowRef<AbortController | null>(null);
+
+  // 增量节流：累积到 pending，rAF 合并刷新一次，避免逐 token reflow。
+  const pending = new Map<string, string>(); // agentId -> 待追加文本
+  let rafScheduled = false;
+  const bubbleByAgent = new Map<string, number>(); // agentId -> dialogue.id
+
+  function flushPending() {
+    rafScheduled = false;
+    if (pending.size === 0) return;
+    for (const [agentId, text] of pending) {
+      const bid = bubbleByAgent.get(agentId);
+      const d = dialogues.value.find((x) => x.id === bid);
+      if (d) d.content += text;
+    }
+    pending.clear();
+    // 触发数组响应（对象内字段已变更，这里强制一次浅刷新以更新双列视图）。
+    dialogues.value = [...dialogues.value];
+  }
+
+  function scheduleFlush() {
+    if (rafScheduled) return;
+    rafScheduled = true;
+    requestAnimationFrame(flushPending);
+  }
+
+  function setSeatStatus(agentId: string, status: SeatView["status"]) {
+    const s = seats.value.find((x) => x.agentId === agentId);
+    if (s) s.status = status;
+    seats.value = [...seats.value];
+  }
+
+  function findDialogue(agentId: string): Dialogue | undefined {
+    const bid = bubbleByAgent.get(agentId);
+    return dialogues.value.find((x) => x.id === bid);
+  }
+
+  function handleEvent(ev: StreamEvent) {
+    switch (ev.type) {
+      case "agent_start":
+        setSeatStatus(ev.agent_id, "running");
+        break;
+      case "delta":
+        pending.set(ev.agent_id, (pending.get(ev.agent_id) ?? "") + ev.text);
+        scheduleFlush();
+        break;
+      case "step": {
+        const d = findDialogue(ev.agent_id);
+        if (d) {
+          d.steps = d.steps ?? [];
+          d.steps.push({
+            id: `${d.id}-${d.steps.length}`,
+            key: ev.key,
+            label: ev.label,
+            ts: nowTime(),
+            args: ev.args ?? null,
+          });
+          dialogues.value = [...dialogues.value];
+        }
+        break;
+      }
+      case "agent_done": {
+        flushPending();
+        const d = findDialogue(ev.agent_id);
+        if (d) {
+          if (ev.content) d.content = ev.content; // 终态正文以后端为准
+          d.streaming = false;
+          d.confidence = "已完成";
+        }
+        setSeatStatus(ev.agent_id, "done");
+        dialogues.value = [...dialogues.value];
+        break;
+      }
+      case "agent_error": {
+        flushPending();
+        const d = findDialogue(ev.agent_id);
+        if (d) {
+          d.content = d.content || `⚠️ 该智能体执行出错：${ev.message}`;
+          d.streaming = false;
+          d.confidence = "出错";
+        }
+        setSeatStatus(ev.agent_id, "error");
+        dialogues.value = [...dialogues.value];
+        break;
+      }
+      case "all_done":
+        flushPending();
+        break;
+    }
+  }
+
+  /** 启动一轮并行运行。会清空旧对话、为每个 agent 建一个流式气泡。 */
+  async function start(
+    cfg: ApiConfig,
+    args: { taskName: string; brief: string; agents: AgentInput[]; model?: string },
+  ): Promise<void> {
+    if (running.value) return;
+    reset();
+    running.value = true;
+
+    bubbleByAgent.clear();
+    const newSeats: SeatView[] = [];
+    const newDialogues: Dialogue[] = [];
+    args.agents.forEach((a, i) => {
+      const avatarType = a.avatarType || getAvatarTypeFor(a.agent_id, i);
+      const bubbleId = nextId();
+      bubbleByAgent.set(a.agent_id, bubbleId);
+      newDialogues.push({
+        id: bubbleId,
+        agentId: a.agent_id,
+        sender: a.name || a.agent_id,
+        avatarType,
+        content: "",
+        time: nowTime(),
+        confidence: "并行执行中",
+        streaming: true,
+        steps: [],
+      });
+      newSeats.push({ agentId: a.agent_id, name: a.name || a.agent_id, avatarType, status: "pending", bubbleId });
+    });
+    dialogues.value = newDialogues;
+    seats.value = newSeats;
+
+    const ac = new AbortController();
+    abortRef.value = ac;
+    try {
+      await streamParallelAgents(
+        cfg,
+        {
+          task_name: args.taskName,
+          brief: args.brief,
+          agent_ids: args.agents.map((a) => a.agent_id),
+          model: args.model,
+        },
+        { onEvent: handleEvent, onError: (err) => console.error("[parallel-run]", err) },
+        ac.signal,
+      );
+    } finally {
+      // 收尾：任何仍在 streaming 的气泡都置完成态。
+      for (const d of dialogues.value) {
+        if (d.streaming) {
+          d.streaming = false;
+          if (d.confidence === "并行执行中") d.confidence = "已完成";
+        }
+      }
+      dialogues.value = [...dialogues.value];
+      running.value = false;
+      abortRef.value = null;
+    }
+  }
+
+  function cancel() {
+    abortRef.value?.abort();
+    running.value = false;
+  }
+
+  /** 回显：从保存的对话列表重建（并据此还原席位条为已完成）。 */
+  function hydrate(saved: Dialogue[]) {
+    cancel();
+    bubbleByAgent.clear();
+    const restored = saved.map((d) => ({ ...d, streaming: false }));
+    dialogues.value = restored;
+    const seatMap = new Map<string, SeatView>();
+    restored.forEach((d) => {
+      bubbleByAgent.set(d.agentId, d.id);
+      if (!seatMap.has(d.agentId)) {
+        seatMap.set(d.agentId, {
+          agentId: d.agentId,
+          name: d.sender,
+          avatarType: d.avatarType,
+          status: d.confidence === "出错" ? "error" : "done",
+          bubbleId: d.id,
+        });
+      }
+    });
+    seats.value = [...seatMap.values()];
+  }
+
+  function reset() {
+    cancel();
+    pending.clear();
+    bubbleByAgent.clear();
+    dialogues.value = [];
+    seats.value = [];
+  }
+
+  return { dialogues, seats, running, start, cancel, hydrate, reset };
+}
@@ -0,0 +1,16 @@
+/**
+ * 前端配置：后端接口地址等。嵌入方一般直接改这里的默认值即可；也可在 dev 用
+ * Vite 环境变量覆盖（VITE_PA_BASE_URL）。组件 props 传入时优先级最高。
+ */
+
+/** DeerFlow 网关默认地址（接口都打到这里）。 */
+export const DEFAULT_BASE_URL: string =
+  (import.meta as unknown as { env?: Record<string, string> }).env?.VITE_PA_BASE_URL ||
+  "http://localhost:8001";
+
+/** 可选默认模型名（不填由后端回退到 config.yaml models[0]）。 */
+export const DEFAULT_MODEL: string | undefined =
+  (import.meta as unknown as { env?: Record<string, string> }).env?.VITE_PA_MODEL || undefined;
+
+/** 免密「用户名登录」端点（公共 + CSRF 豁免）。 */
+export const USERNAME_LOGIN_PATH = "/api/v1/auth/login/username";
@@ -0,0 +1,18 @@
+/**
+ * 组件库入口。其它 Vue 项目：
+ *   import { ParallelAgentsPanel } from "parallel-agents-vue";
+ *   import "parallel-agents-vue/style.css";
+ * 或全局注册：app.use(ParallelAgentsVue)。
+ */
+import "./styles/parallel-agents.css";
+import type { App } from "vue";
+import ParallelAgentsPanel from "./components/ParallelAgentsPanel.vue";
+
+export { ParallelAgentsPanel };
+export type { AgentInput, Dialogue, SavedTask, SeatView, StreamEvent } from "./types";
+
+export default {
+  install(app: App) {
+    app.component("ParallelAgentsPanel", ParallelAgentsPanel);
+  },
+};
@@ -0,0 +1,27 @@
+/** 头像配色（对齐圆桌 Step2 的 avatarType：blue/purple/emerald/amber/pink/human）。 */
+
+export const FALLBACK_AVATAR_TYPES = ["blue", "purple", "emerald", "amber", "pink"] as const;
+
+/** 无显式 avatarType 时按下标轮转一个配色（与 Step2 getAvatarTypeFor 行为一致）。 */
+export function getAvatarTypeFor(_agentId: string, index: number): string {
+  return FALLBACK_AVATAR_TYPES[index % FALLBACK_AVATAR_TYPES.length];
+}
+
+/** 每种 avatarType 在深色面板里的底色 + 主体色（机器人头 / 真人）。 */
+export const AVATAR_COLORS: Record<string, { bg: string; fg: string }> = {
+  blue: { bg: "rgba(22,119,255,0.18)", fg: "#4096ff" },
+  purple: { bg: "rgba(124,58,237,0.20)", fg: "#a78bfa" },
+  emerald: { bg: "rgba(16,185,129,0.20)", fg: "#34d399" },
+  amber: { bg: "rgba(245,158,11,0.20)", fg: "#fbbf24" },
+  pink: { bg: "rgba(219,39,119,0.20)", fg: "#f472b6" },
+  human: { bg: "rgba(148,163,184,0.20)", fg: "#cbd5e1" },
+  man: { bg: "rgba(148,163,184,0.20)", fg: "#cbd5e1" },
+};
+
+export function avatarColor(type: string): { bg: string; fg: string } {
+  return AVATAR_COLORS[type] ?? AVATAR_COLORS.blue;
+}
+
+export function isHuman(type: string): boolean {
+  return type === "human" || type === "man";
+}
@@ -0,0 +1,5 @@
+import { createApp } from "vue";
+import App from "./App.vue";
+import "./styles/parallel-agents.css";
+
+createApp(App).mount("#app");
@@ -0,0 +1,282 @@
+/* 并行多智能体面板样式。
+ *
+ * 嵌入安全：只引 Tailwind 的 theme + utilities（**不引 preflight**，避免重置宿主全局样式）。
+ * 颜色用作用域内的 `--pa-*` 深蓝 token（取自《深蓝 UI 规范》ZS-004 / ZS-002，与圆桌 Step2
+ * 深色一致），全部包在根类 `.parallel-agents` 下，绝不污染宿主。布局用 Tailwind 工具类，
+ * 主题色用下面的 `.pa-*` 语义类。 */
+
+@layer theme, base, components, utilities;
+@import "tailwindcss/theme.css" layer(theme);
+@import "tailwindcss/utilities.css" layer(utilities);
+
+/* ── 作用域 token（深色深蓝，自带成色，不依赖宿主主题）─────────────────────── */
+.parallel-agents {
+  --pa-bg: #0c1836; /* 页面背景 zs-004-09 */
+  --pa-card: #122347; /* 卡片 / 气泡容器 zs-004-08 */
+  --pa-muted: #16294d; /* 次级表面（步骤卡 / 并行列底） */
+  --pa-accent: #1a2f5a; /* hover / 软高亮 */
+  --pa-fg: #c5cfe0; /* 正文 zs-004-02 */
+  --pa-fg-muted: #6e88b0; /* 次级文字 zs-004-04 */
+  --pa-border: #2d4473; /* 容器边框 zs-004-06 */
+  --pa-input: #4a6491; /* 控件描边 */
+  --pa-primary: #1677ff; /* 主操作 zs-002-06 */
+  --pa-primary-fg: #ffffff;
+  --pa-chip-bg: #16335f; /* badge 底 */
+  --pa-chip-fg: #4096ff; /* badge 文字 */
+  --pa-amber: #f59e0b;
+  --pa-emerald: #10b981;
+  --pa-rose: #ef4444;
+
+  color: var(--pa-fg);
+  background-color: var(--pa-bg);
+  font-size: 14px;
+  line-height: 1.6;
+  box-sizing: border-box;
+}
+.parallel-agents *,
+.parallel-agents *::before,
+.parallel-agents *::after {
+  box-sizing: border-box;
+}
+
+/* 滚动条 */
+.parallel-agents ::-webkit-scrollbar {
+  width: 8px;
+  height: 8px;
+}
+.parallel-agents ::-webkit-scrollbar-thumb {
+  background: color-mix(in oklch, var(--pa-primary) 40%, var(--pa-border));
+  border-radius: 999px;
+}
+.parallel-agents ::-webkit-scrollbar-thumb:hover {
+  background: color-mix(in oklch, var(--pa-primary) 60%, var(--pa-border));
+}
+
+/* ── 语义颜色类（组件用这些上色，布局仍用 Tailwind 工具类）──────────────────── */
+.parallel-agents .pa-bg {
+  background-color: var(--pa-bg);
+}
+.parallel-agents .pa-card {
+  background-color: var(--pa-card);
+}
+.parallel-agents .pa-surface {
+  background-color: var(--pa-muted);
+}
+.parallel-agents .pa-surface-soft {
+  background-color: color-mix(in oklch, var(--pa-muted) 55%, transparent);
+}
+.parallel-agents .pa-border {
+  border-color: var(--pa-border);
+}
+.parallel-agents .pa-text {
+  color: var(--pa-fg);
+}
+.parallel-agents .pa-text-muted {
+  color: var(--pa-fg-muted);
+}
+.parallel-agents .pa-text-primary {
+  color: var(--pa-primary);
+}
+
+/* 输入框 / 文本域 */
+.parallel-agents .pa-input {
+  background-color: var(--pa-card);
+  border: 1px solid var(--pa-border);
+  color: var(--pa-fg);
+  border-radius: 8px;
+}
+.parallel-agents .pa-input::placeholder {
+  color: color-mix(in oklch, var(--pa-fg-muted) 80%, transparent);
+}
+.parallel-agents .pa-input:focus {
+  outline: none;
+  border-color: color-mix(in oklch, var(--pa-primary) 55%, var(--pa-border));
+}
+
+/* 主按钮 / 次按钮 */
+.parallel-agents .pa-btn-primary {
+  background-color: var(--pa-primary);
+  color: var(--pa-primary-fg);
+  border-radius: 8px;
+  transition: filter 0.15s ease;
+}
+.parallel-agents .pa-btn-primary:hover {
+  filter: brightness(1.08);
+}
+.parallel-agents .pa-btn-primary:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+.parallel-agents .pa-btn-ghost {
+  border: 1px solid var(--pa-border);
+  color: var(--pa-fg);
+  background-color: transparent;
+  border-radius: 8px;
+  transition: background-color 0.15s ease;
+}
+.parallel-agents .pa-btn-ghost:hover {
+  background-color: var(--pa-accent);
+}
+.parallel-agents .pa-btn-ghost:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+
+/* 状态徽标（"实时输出" / "已完成" …）—— 对齐 Step2 confidence chip */
+.parallel-agents .pa-badge {
+  background-color: var(--pa-chip-bg);
+  color: var(--pa-chip-fg);
+  border: 1px solid var(--pa-border);
+  font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+  font-weight: 800;
+  text-transform: uppercase;
+  border-radius: 4px;
+  padding: 1px 6px;
+  font-size: 11px;
+  letter-spacing: 0.02em;
+}
+
+/* 席位状态角标色 */
+.parallel-agents .pa-dot-running {
+  background-color: var(--pa-primary);
+  color: #fff;
+}
+.parallel-agents .pa-dot-done {
+  background-color: var(--pa-emerald);
+  color: #fff;
+}
+.parallel-agents .pa-dot-error {
+  background-color: var(--pa-rose);
+  color: #fff;
+}
+
+/* 流式光标 + loading 点 */
+.parallel-agents .pa-cursor {
+  display: inline-block;
+  width: 4px;
+  height: 13px;
+  margin-left: 2px;
+  vertical-align: middle;
+  background: var(--pa-primary);
+  animation: pa-pulse 1s ease-in-out infinite;
+}
+@keyframes pa-pulse {
+  0%,
+  100% {
+    opacity: 1;
+  }
+  50% {
+    opacity: 0.25;
+  }
+}
+.parallel-agents .pa-bounce {
+  animation: pa-bounce 0.9s infinite;
+}
+@keyframes pa-bounce {
+  0%,
+  100% {
+    transform: translateY(0);
+  }
+  50% {
+    transform: translateY(-3px);
+  }
+}
+@keyframes pa-spin {
+  to {
+    transform: rotate(360deg);
+  }
+}
+.parallel-agents .pa-spin {
+  animation: pa-spin 0.8s linear infinite;
+}
+@keyframes pa-ping {
+  75%,
+  100% {
+    transform: scale(1.8);
+    opacity: 0;
+  }
+}
+.parallel-agents .pa-ping {
+  animation: pa-ping 1.2s cubic-bezier(0, 0, 0.2, 1) infinite;
+}
+
+/* ── Markdown 正文（对齐 Step2 sharedMarkdownComponents 的深色观感）──────────── */
+.parallel-agents .pa-md {
+  color: var(--pa-fg);
+  font-size: 14px;
+  line-height: 1.7;
+  word-break: break-word;
+}
+.parallel-agents .pa-md p {
+  margin: 0.35em 0;
+}
+.parallel-agents .pa-md h1,
+.parallel-agents .pa-md h2,
+.parallel-agents .pa-md h3,
+.parallel-agents .pa-md h4 {
+  font-weight: 700;
+  margin: 0.7em 0 0.35em;
+  color: var(--pa-fg);
+}
+.parallel-agents .pa-md h1 {
+  font-size: 1.3em;
+}
+.parallel-agents .pa-md h2 {
+  font-size: 1.18em;
+}
+.parallel-agents .pa-md h3 {
+  font-size: 1.06em;
+}
+.parallel-agents .pa-md ul,
+.parallel-agents .pa-md ol {
+  margin: 0.35em 0;
+  padding-left: 1.4em;
+}
+.parallel-agents .pa-md li {
+  margin: 0.2em 0;
+}
+.parallel-agents .pa-md a {
+  color: var(--pa-chip-fg);
+  text-decoration: underline;
+}
+.parallel-agents .pa-md code {
+  background: var(--pa-muted);
+  border: 1px solid var(--pa-border);
+  border-radius: 4px;
+  padding: 0.1em 0.35em;
+  font-size: 0.9em;
+  font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+}
+.parallel-agents .pa-md pre {
+  background: var(--pa-muted);
+  border: 1px solid var(--pa-border);
+  border-radius: 8px;
+  padding: 0.75em;
+  overflow-x: auto;
+  margin: 0.5em 0;
+}
+.parallel-agents .pa-md pre code {
+  background: transparent;
+  border: 0;
+  padding: 0;
+}
+.parallel-agents .pa-md blockquote {
+  border-left: 3px solid var(--pa-border);
+  padding-left: 0.75em;
+  color: var(--pa-fg-muted);
+  margin: 0.5em 0;
+}
+.parallel-agents .pa-md table {
+  border-collapse: collapse;
+  margin: 0.5em 0;
+  width: 100%;
+}
+.parallel-agents .pa-md th,
+.parallel-agents .pa-md td {
+  border: 1px solid var(--pa-border);
+  padding: 0.35em 0.6em;
+  text-align: left;
+}
+.parallel-agents .pa-md th {
+  background: var(--pa-muted);
+}
@@ -0,0 +1,67 @@
+/**
+ * 并行多智能体面板的共享类型。
+ *
+ * `Dialogue` 镜像圆桌 Step2 的 `Step2Dialogue`，外加 `agentId`（并行场景下用来把消息
+ * 归到对应席位列）。`StreamEvent` 是后端 `/api/parallel-agents/run/stream` 的统一事件协议。
+ */
+
+/** 一条工具调用步骤（喂步骤条 / 思维链卡）。 */
+export interface DialogueStep {
+  id: string;
+  key: string;
+  label: string;
+  ts: string;
+  args?: Record<string, unknown> | null;
+}
+
+/** 一条消息气泡（对齐 Step2Dialogue + agentId）。 */
+export interface Dialogue {
+  id: number;
+  /** 归属席位的 agent_id（并行列定位用）。 */
+  agentId: string;
+  sender: string;
+  avatarType: string;
+  content: string;
+  time: string;
+  /** 状态标签（"实时输出" / "已完成" / "出错" 等）。 */
+  confidence: string;
+  streaming?: boolean;
+  steps?: DialogueStep[];
+}
+
+/** 并行席位的视图状态（席位状态条用）。 */
+export interface SeatView {
+  agentId: string;
+  name: string;
+  avatarType: string;
+  status: "pending" | "running" | "done" | "error";
+  /** 关联的 Dialogue.id。 */
+  bubbleId: number;
+}
+
+/** 宿主传入的一个待并行调用的智能体。 */
+export interface AgentInput {
+  agent_id: string;
+  name: string;
+  avatarType?: string;
+}
+
+/** 后端合并 SSE 的统一事件。 */
+export type StreamEvent =
+  | { type: "agent_start"; agent_id: string }
+  | { type: "delta"; agent_id: string; text: string }
+  | { type: "step"; agent_id: string; key: string; label: string; args?: Record<string, unknown> | null }
+  | { type: "agent_done"; agent_id: string; content: string }
+  | { type: "agent_error"; agent_id: string; message: string }
+  | { type: "all_done" };
+
+/** 持久化的任务记录（与后端 /api/parallel-agent-tasks 对应）。 */
+export interface SavedTask {
+  id: string;
+  title: string;
+  brief?: string | null;
+  agents: AgentInput[];
+  messages?: Dialogue[];
+  created_at?: string | null;
+  updated_at?: string | null;
+}
@@ -0,0 +1,7 @@
+/// <reference types="vite/client" />
+
+declare module "*.vue" {
+  import type { DefineComponent } from "vue";
+  const component: DefineComponent<{}, {}, any>;
+  export default component;
+}
@@ -0,0 +1,23 @@
+{
+  "compilerOptions": {
+    "target": "ES2020",
+    "useDefineForClassFields": true,
+    "module": "ESNext",
+    "lib": ["ES2020", "DOM", "DOM.Iterable"],
+    "skipLibCheck": true,
+    "moduleResolution": "bundler",
+    "allowImportingTsExtensions": true,
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "noEmit": true,
+    "jsx": "preserve",
+    "strict": true,
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "noFallthroughCasesInSwitch": true,
+    "types": ["vite/client"],
+    "baseUrl": ".",
+    "paths": { "@/*": ["./src/*"] }
+  },
+  "include": ["src/**/*.ts", "src/**/*.d.ts", "src/**/*.vue"]
+}
@@ -0,0 +1,40 @@
+import { fileURLToPath, URL } from "node:url";
+import { defineConfig } from "vite";
+import vue from "@vitejs/plugin-vue";
+import tailwindcss from "@tailwindcss/vite";
+
+// 两种构建：
+//  - 默认（`vite build` / `vite dev`）：独立 demo 应用（index.html）。
+//  - `--mode lib`（`pnpm build:lib`）：组件库，产出 ESM + UMD + 一个 style.css，
+//    供其它 Vue 项目 `import { ParallelAgentsPanel } from "parallel-agents-vue"`。
+export default defineConfig(({ mode }) => {
+  const isLib = mode === "lib";
+  return {
+    plugins: [vue(), tailwindcss()],
+    resolve: {
+      alias: { "@": fileURLToPath(new URL("./src", import.meta.url)) },
+    },
+    server: { port: 5183 },
+    build: isLib
+      ? {
+          lib: {
+            entry: fileURLToPath(new URL("./src/index.ts", import.meta.url)),
+            name: "ParallelAgentsVue",
+            fileName: "parallel-agents-vue",
+          },
+          rollupOptions: {
+            // vue 由宿主提供，不打进库。
+            external: ["vue"],
+            output: {
+              exports: "named",
+              globals: { vue: "Vue" },
+              assetFileNames: "parallel-agents-vue.[ext]",
+            },
+          },
+          cssCodeSplit: false,
+        }
+      : {
+          outDir: "dist-app",
+        },
+  };
+});