a924486f26
- Create suyao_app_service.py and suyao_ws_handler.py for 苏瑶 Feishu bot
- Add SUYAO_APP_ID/SUYAO_APP_SECRET/SUYAO_AGENT_ID config fields
- Fix node config extraction bug (n.get("config") → n.get("data")) in all WS handlers
- Add _build_memory_config_from_node() to support per-agent memory settings
(max_history_messages, vector_memory_top_k, persist_to_db, etc.)
- Create 苏瑶1号 (Plan A: long context), 苏瑶2号 (Plan B: emotion tracking),
苏瑶3号 (Plan C: knowledge graph + RAG) agents with different memory strategies
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
419 lines
15 KiB
Python
419 lines
15 KiB
Python
"""
|
||
Agent 独立聊天 API — 不依赖工作流 DAG，直接与 Agent Runtime 对话。
|
||
|
||
POST /api/v1/agent-chat/bare
|
||
{"message": "你好，帮我..."}
|
||
→ {"content": "...", "iterations": 3, "tool_calls": 5}
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import logging
|
||
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
|
||
from sqlalchemy.orm import Session
|
||
from app.api.auth import get_current_user
|
||
from app.models.user import User
|
||
from app.models.agent import Agent
|
||
from app.models.agent_llm_log import AgentLLMLog
|
||
from app.agent_runtime import (
|
||
AgentRuntime,
|
||
AgentConfig,
|
||
AgentLLMConfig,
|
||
AgentToolConfig,
|
||
AgentBudgetConfig,
|
||
AgentMemoryConfig,
|
||
AgentStep,
|
||
AgentOrchestrator,
|
||
OrchestratorAgentConfig,
|
||
)
|
||
from app.core.config import settings
|
||
|
||
logger = logging.getLogger(__name__)
|
||
router = APIRouter(prefix="/api/v1/agent-chat", tags=["agent-chat"])
|
||
|
||
|
||
def _make_llm_logger(
|
||
db: Session,
|
||
agent_id: Optional[str] = None,
|
||
user_id: Optional[str] = None,
|
||
):
|
||
"""创建 LLM 调用日志回调，写入 AgentLLMLog 表。"""
|
||
def _log(metrics: dict):
|
||
try:
|
||
log = AgentLLMLog(
|
||
agent_id=agent_id,
|
||
session_id=metrics.get("session_id"),
|
||
user_id=user_id,
|
||
model=metrics.get("model", ""),
|
||
provider=metrics.get("provider"),
|
||
prompt_tokens=metrics.get("prompt_tokens", 0),
|
||
completion_tokens=metrics.get("completion_tokens", 0),
|
||
total_tokens=metrics.get("total_tokens", 0),
|
||
latency_ms=metrics.get("latency_ms", 0),
|
||
iteration_number=metrics.get("iteration_number", 0),
|
||
step_type=metrics.get("step_type"),
|
||
tool_name=metrics.get("tool_name"),
|
||
status=metrics.get("status", "success"),
|
||
error_message=metrics.get("error_message"),
|
||
)
|
||
db.add(log)
|
||
db.commit()
|
||
except Exception as e:
|
||
logger.warning("写入 AgentLLMLog 失败: %s", e)
|
||
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
|
||
model: Optional[str] = None
|
||
temperature: Optional[float] = None
|
||
max_iterations: Optional[int] = None
|
||
|
||
|
||
class ChatResponse(BaseModel):
|
||
content: str
|
||
iterations_used: int
|
||
tool_calls_made: int
|
||
truncated: bool
|
||
session_id: str
|
||
agent_id: Optional[str] = None
|
||
steps: List[AgentStep] = Field(default_factory=list, description="执行追踪步骤")
|
||
|
||
|
||
class OrchestrateAgentItem(BaseModel):
|
||
"""编排中单个 Agent 的定义"""
|
||
id: str
|
||
name: str = "Agent"
|
||
system_prompt: str = "你是一个有用的AI助手。"
|
||
model: str = "deepseek-v4-flash"
|
||
provider: str = "deepseek"
|
||
temperature: float = 0.7
|
||
max_iterations: int = 10
|
||
tools: List[str] = Field(default_factory=list)
|
||
description: str = ""
|
||
|
||
|
||
class OrchestrateRequest(BaseModel):
|
||
"""多 Agent 编排请求"""
|
||
message: str
|
||
mode: str = "debate"
|
||
agents: List[OrchestrateAgentItem] = Field(..., min_length=1)
|
||
model: Optional[str] = None
|
||
|
||
|
||
class OrchestrateStepItem(BaseModel):
|
||
"""编排步骤"""
|
||
agent_id: str
|
||
agent_name: str
|
||
input: str = ""
|
||
output: str = ""
|
||
iterations_used: int = 0
|
||
tool_calls_made: int = 0
|
||
error: Optional[str] = None
|
||
|
||
|
||
class OrchestrateResponse(BaseModel):
|
||
"""多 Agent 编排响应"""
|
||
mode: str
|
||
final_answer: str
|
||
steps: List[OrchestrateStepItem] = Field(default_factory=list)
|
||
agent_results: List[Dict[str, Any]] = Field(default_factory=list)
|
||
|
||
|
||
@router.post("/orchestrate", response_model=OrchestrateResponse)
|
||
async def orchestrate_agents(
|
||
req: OrchestrateRequest,
|
||
current_user: User = Depends(get_current_user),
|
||
db: Session = Depends(get_db),
|
||
):
|
||
"""多 Agent 编排：支持 route / sequential / debate 三种模式。"""
|
||
agents = [
|
||
OrchestratorAgentConfig(
|
||
id=a.id, name=a.name,
|
||
system_prompt=a.system_prompt,
|
||
model=req.model or a.model,
|
||
provider=a.provider,
|
||
temperature=a.temperature,
|
||
max_iterations=a.max_iterations,
|
||
tools=a.tools,
|
||
description=a.description,
|
||
)
|
||
for a in req.agents
|
||
]
|
||
|
||
on_llm_call = _make_llm_logger(db, agent_id=None, user_id=current_user.id)
|
||
orchestrator = AgentOrchestrator(
|
||
default_llm_config=AgentLLMConfig(
|
||
model=req.model or "deepseek-v4-flash",
|
||
temperature=0.3,
|
||
),
|
||
)
|
||
result = await orchestrator.run(req.mode, req.message, agents, on_llm_call=on_llm_call)
|
||
|
||
return OrchestrateResponse(
|
||
mode=result.mode,
|
||
final_answer=result.final_answer,
|
||
steps=[
|
||
OrchestrateStepItem(
|
||
agent_id=s.agent_id,
|
||
agent_name=s.agent_name,
|
||
input=s.input,
|
||
output=s.output,
|
||
iterations_used=s.iterations_used,
|
||
tool_calls_made=s.tool_calls_made,
|
||
error=s.error,
|
||
)
|
||
for s in result.steps
|
||
],
|
||
agent_results=result.agent_results,
|
||
)
|
||
|
||
|
||
@router.post("/bare", response_model=ChatResponse)
|
||
async def chat_bare(
|
||
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。"""
|
||
uid = current_user.id
|
||
bare_scope = f"{uid}:__bare__" if uid else "__bare__"
|
||
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=uid,
|
||
memory_scope_id=bare_scope,
|
||
)
|
||
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)
|
||
result = await runtime.run(req.message)
|
||
|
||
return ChatResponse(
|
||
content=result.content,
|
||
iterations_used=result.iterations_used,
|
||
tool_calls_made=result.tool_calls_made,
|
||
truncated=result.truncated,
|
||
session_id=runtime.context.session_id,
|
||
steps=result.steps,
|
||
)
|
||
|
||
|
||
@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）。"""
|
||
uid = current_user.id
|
||
bare_scope = f"{uid}:__bare__" if uid else "__bare__"
|
||
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=uid,
|
||
memory_scope_id=bare_scope,
|
||
)
|
||
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,
|
||
req: ChatRequest,
|
||
current_user: User = Depends(get_current_user),
|
||
db: Session = Depends(get_db),
|
||
):
|
||
"""与指定的 Agent 对话。Agent 的工作流配置会用于构建 Runtime。"""
|
||
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")
|
||
|
||
# 从 Agent 配置构建 Runtime
|
||
wc = agent.workflow_config or {}
|
||
nodes = wc.get("nodes", [])
|
||
# 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"]))
|
||
|
||
uid = current_user.id
|
||
mem_scope = f"{uid}:{agent_id}" if uid else str(agent_id)
|
||
memory_cfg = _build_memory_config_from_node(agent_node_cfg)
|
||
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", []),
|
||
),
|
||
memory=memory_cfg,
|
||
budget=budget,
|
||
user_id=uid,
|
||
memory_scope_id=mem_scope,
|
||
)
|
||
|
||
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)
|
||
result = await runtime.run(req.message)
|
||
|
||
return ChatResponse(
|
||
content=result.content,
|
||
iterations_used=result.iterations_used,
|
||
tool_calls_made=result.tool_calls_made,
|
||
truncated=result.truncated,
|
||
session_id=runtime.context.session_id,
|
||
agent_id=agent_id,
|
||
steps=result.steps,
|
||
)
|
||
|
||
|
||
@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"]))
|
||
|
||
uid = current_user.id
|
||
mem_scope = f"{uid}:{agent_id}" if uid else str(agent_id)
|
||
memory_cfg = _build_memory_config_from_node(agent_node_cfg)
|
||
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", []),
|
||
),
|
||
memory=memory_cfg,
|
||
budget=budget,
|
||
user_id=uid,
|
||
memory_scope_id=mem_scope,
|
||
)
|
||
|
||
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:
|
||
return {}
|
||
for node in nodes:
|
||
typ = node.get("type", "")
|
||
if typ in ("agent", "llm", "template"):
|
||
return node.get("data") or {}
|
||
return {}
|
||
|
||
|
||
def _build_memory_config_from_node(agent_node_cfg: dict) -> AgentMemoryConfig:
|
||
"""从 Agent 工作流节点配置中提取记忆配置。"""
|
||
return AgentMemoryConfig(
|
||
max_history_messages=int(agent_node_cfg.get("memory_max_history", 20)),
|
||
vector_memory_top_k=int(agent_node_cfg.get("memory_vector_top_k", 5)),
|
||
persist_to_db=bool(agent_node_cfg.get("memory_persist", True)),
|
||
vector_memory_enabled=bool(agent_node_cfg.get("memory_vector_enabled", True)),
|
||
learning_enabled=bool(agent_node_cfg.get("memory_learning", True)),
|
||
)
|