feat: 向量记忆 RAG、工具市场、SSE 流式响应、前端集成与测试覆盖

- 新增 embedding_service（语义检索）、knowledge_service（RAG）、text_chunker、document_parser - 新增 tool_registry（自定义工具注册表）并完善工具市场 API（CRUD + code/http 执行） - 新增 agent_vector_memory / knowledge_base 模型及对应数据库表 - 实现 SSE 流式响应与 Agent 预算控制 - AgentChat.vue 集成 MainLayout 导航布局 - 完善测试体系：7 个新测试文件共 110 个测试覆盖 - 修复 conftest.py SQLite 内存数据库连接隔离问题 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-01 22:30:46 +08:00
parent 036f533881
commit 7b9e0826de
35 changed files with 4353 additions and 365 deletions
--- a/backend/app/api/agent_chat.py
+++ b/backend/app/api/agent_chat.py
@@ -8,8 +8,10 @@ POST /api/v1/agent-chat/bare
 from __future__ import annotations

 import logging
-from typing import Any, Dict, List, Optional
-from fastapi import APIRouter, Depends, HTTPException
+import json
+from typing import Any, AsyncGenerator, Dict, List, Optional
+from fastapi import APIRouter, Depends, HTTPException, Request
+from fastapi.responses import StreamingResponse
 from pydantic import BaseModel, Field

 from app.core.database import get_db
@@ -23,6 +25,7 @@ from app.agent_runtime import (
    AgentConfig,
    AgentLLMConfig,
    AgentToolConfig,
+    AgentBudgetConfig,
    AgentStep,
    AgentOrchestrator,
    OrchestratorAgentConfig,
@@ -64,6 +67,14 @@ def _make_llm_logger(
    return _log


+async def _sse_stream(gen: AsyncGenerator[dict, None]) -> AsyncGenerator[str, None]:
+    """将 run_stream 生成的 dict 事件格式化为 SSE 文本流。"""
+    async for event in gen:
+        event_type = event.get("type", "message")
+        data = {k: v for k, v in event.items() if k != "type"}
+        yield f"event: {event_type}\ndata: {json.dumps(data, ensure_ascii=False)}\n\n"
+
+
 class ChatRequest(BaseModel):
    message: str
    session_id: Optional[str] = None
@@ -205,6 +216,39 @@ async def chat_bare(
    )


+@router.post("/bare/stream")
+async def chat_bare_stream(
+    req: ChatRequest,
+    current_user: User = Depends(get_current_user),
+    db: Session = Depends(get_db),
+):
+    """无需 Agent 配置，使用默认设置直接对话（流式 SSE）。"""
+    config = AgentConfig(
+        name="bare_agent",
+        system_prompt="你是一个有用的AI助手。请使用可用工具来帮助用户完成任务。",
+        llm=AgentLLMConfig(
+            model=req.model or (
+                "gpt-4o-mini" if settings.OPENAI_API_KEY and settings.OPENAI_API_KEY != "your-openai-api-key"
+                else "deepseek-v4-flash"
+            ),
+            temperature=req.temperature or 0.7,
+            max_iterations=req.max_iterations or 10,
+        ),
+        user_id=current_user.id,
+    )
+    on_llm_call = _make_llm_logger(db, agent_id=None, user_id=current_user.id)
+    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call)
+    return StreamingResponse(
+        _sse_stream(runtime.run_stream(req.message)),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
@router.post("/{agent_id}", response_model=ChatResponse)
 async def chat_with_agent(
    agent_id: str,
@@ -225,9 +269,25 @@ async def chat_with_agent(
    # 查找 agent 节点的配置（或第一个 llm 节点的配置）
    agent_node_cfg = _find_agent_node_config(nodes)

+    # 构建 system prompt，并自动注入智能体名称
+    system_prompt = agent_node_cfg.get("system_prompt") or agent.description or "你是一个有用的AI助手。"
+    if agent.name:
+        name_prefix = f"你的名字是{agent.name}"
+        if name_prefix not in system_prompt:
+            system_prompt = f"{name_prefix}。\n\n{system_prompt}"
+
+    # 合并执行预算：Agent.budget_config 覆盖默认值
+    budget = AgentBudgetConfig()
+    if agent.budget_config and isinstance(agent.budget_config, dict):
+        bc = agent.budget_config
+        if "max_llm_invocations" in bc and bc["max_llm_invocations"] is not None:
+            budget.max_llm_invocations = max(1, int(bc["max_llm_invocations"]))
+        if "max_tool_calls" in bc and bc["max_tool_calls"] is not None:
+            budget.max_tool_calls = max(1, int(bc["max_tool_calls"]))
+
    config = AgentConfig(
        name=agent.name,
-        system_prompt=agent_node_cfg.get("system_prompt") or agent.description or "你是一个有用的AI助手。",
+        system_prompt=system_prompt,
        llm=AgentLLMConfig(
            provider=agent_node_cfg.get("provider", "openai"),
            model=req.model or agent_node_cfg.get("model", "gpt-4o-mini"),
@@ -238,6 +298,7 @@ async def chat_with_agent(
            include_tools=agent_node_cfg.get("tools", []),
            exclude_tools=agent_node_cfg.get("exclude_tools", []),
        ),
+        budget=budget,
        user_id=current_user.id,
    )

@@ -256,6 +317,68 @@ async def chat_with_agent(
    )


+@router.post("/{agent_id}/stream")
+async def chat_with_agent_stream(
+    agent_id: str,
+    req: ChatRequest,
+    current_user: User = Depends(get_current_user),
+    db: Session = Depends(get_db),
+):
+    """与指定的 Agent 对话（流式 SSE）。"""
+    agent = db.query(Agent).filter(Agent.id == agent_id).first()
+    if not agent:
+        raise HTTPException(status_code=404, detail="Agent 不存在")
+    if agent.user_id and agent.user_id != current_user.id and current_user.role != "admin":
+        raise HTTPException(status_code=403, detail="无权访问该 Agent")
+
+    wc = agent.workflow_config or {}
+    nodes = wc.get("nodes", [])
+    agent_node_cfg = _find_agent_node_config(nodes)
+
+    system_prompt = agent_node_cfg.get("system_prompt") or agent.description or "你是一个有用的AI助手。"
+    if agent.name:
+        name_prefix = f"你的名字是{agent.name}"
+        if name_prefix not in system_prompt:
+            system_prompt = f"{name_prefix}。\n\n{system_prompt}"
+
+    budget = AgentBudgetConfig()
+    if agent.budget_config and isinstance(agent.budget_config, dict):
+        bc = agent.budget_config
+        if "max_llm_invocations" in bc and bc["max_llm_invocations"] is not None:
+            budget.max_llm_invocations = max(1, int(bc["max_llm_invocations"]))
+        if "max_tool_calls" in bc and bc["max_tool_calls"] is not None:
+            budget.max_tool_calls = max(1, int(bc["max_tool_calls"]))
+
+    config = AgentConfig(
+        name=agent.name,
+        system_prompt=system_prompt,
+        llm=AgentLLMConfig(
+            provider=agent_node_cfg.get("provider", "openai"),
+            model=req.model or agent_node_cfg.get("model", "gpt-4o-mini"),
+            temperature=req.temperature or float(agent_node_cfg.get("temperature", 0.7)),
+            max_iterations=req.max_iterations or int(agent_node_cfg.get("max_iterations", 10)),
+        ),
+        tools=AgentToolConfig(
+            include_tools=agent_node_cfg.get("tools", []),
+            exclude_tools=agent_node_cfg.get("exclude_tools", []),
+        ),
+        budget=budget,
+        user_id=current_user.id,
+    )
+
+    on_llm_call = _make_llm_logger(db, agent_id=agent_id, user_id=current_user.id)
+    runtime = AgentRuntime(config=config, on_llm_call=on_llm_call)
+    return StreamingResponse(
+        _sse_stream(runtime.run_stream(req.message)),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
 def _find_agent_node_config(nodes: list) -> Dict[str, Any]:
    """从工作流节点列表中查找第一个 agent 类型或 llm 类型的节点配置。"""
    if not nodes: