fix: chat stream 走 regulatory agent 支持工具调用,修复 workflow ValidationError

1. chat.py stream 端点改为调用 regulatory_node.stream_working()(pydantic-ai
   run_stream),支持工具调用 + 逐 token 流式输出
2. regulatory_node 新增 stream_working 方法,通过 asyncio.Queue 推送 token
3. ConsciousnessNodeDeps.available_skills 加默认值 None,修复 ForWorkflowInput/
   ForregulatoryInput 路径的 ValidationError

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
2026-06-05 07:37:59 +00:00
parent 6f1bc27101
commit b61524e5d9
3 changed files with 57 additions and 62 deletions
+24 -61
View File
@@ -14,7 +14,6 @@
import json
import asyncio
import httpx
from fastapi import APIRouter, Depends, HTTPException, Request
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
@@ -164,7 +163,7 @@ async def stream_chat_message(
request: Request,
token_data: TokenData = Depends(Accessor.get_current_user),
):
"""SSE 流式聊天端点:逐 token 推送 AI 回复"""
"""SSE 流式聊天端点:通过 regulatory_node agent 流式输出,支持工具调用"""
postgres_database = ray_actor_hook("postgres_database").postgres_database
# 存用户消息
@@ -172,71 +171,35 @@ async def stream_chat_message(
chat_id=chat_id, message=request_body.message, message_owner="user"
)
# 获取 regulatory_node 的 provider 配置
node_config = await postgres_database.get_system_node_config.remote("regulatory_node")
if not node_config:
raise HTTPException(status_code=500, detail="Regulatory node not configured")
# 构造 MessageRequest payload
payload = MessageRequest(
platform="client",
user_name=token_data.user_id,
platform_id=chat_id,
message=request_body.message,
)
# 获取 provider 详情
from kilostar.core.global_state_machine.gsm_snapshot import fetch_snapshot
regulatory_node = ray_actor_hook("regulatory_node").regulatory_node
token_queue = asyncio.Queue()
global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
snapshot = await fetch_snapshot(gsm_actor=global_state_machine)
provider = snapshot.providers.get(node_config.provider_title)
if not provider:
raise HTTPException(status_code=500, detail="Provider not available")
# 加载历史消息作为上下文
history_msgs = await postgres_database.list_chat_messages.remote(chat_id=chat_id)
messages = []
system_prompt = "你是 KiloStar 助手,友善、简洁地回答用户的问题。"
if node_config.persona_id:
tpl = await postgres_database.get_template.remote(node_config.persona_id)
if tpl and tpl.system_prompt:
system_prompt += "\n" + tpl.system_prompt
messages.append({"role": "system", "content": system_prompt})
for msg in history_msgs:
role = "user" if msg.message_owner == "user" else "assistant"
messages.append({"role": role, "content": msg.message})
# stream_working.remote() returns an asyncio.Task in standalone mode
stream_task = regulatory_node.stream_working.remote(payload, token_queue)
async def event_generator():
full_response = ""
try:
async with httpx.AsyncClient(timeout=120.0) as client:
url = provider.provider_url.rstrip("/") + "/chat/completions"
payload = {
"model": node_config.model_id,
"messages": messages,
"stream": True,
}
async with client.stream(
"POST",
url,
json=payload,
headers={
"Authorization": f"Bearer {provider.provider_apikey}",
"Content-Type": "application/json",
},
) as resp:
async for line in resp.aiter_lines():
if await request.is_disconnected():
break
if not line.startswith("data: "):
continue
data_str = line[6:]
if data_str.strip() == "[DONE]":
break
try:
chunk = json.loads(data_str)
delta = chunk.get("choices", [{}])[0].get("delta", {})
token = delta.get("content", "")
if token:
full_response += token
yield f"data: {json.dumps({'token': token})}\n\n"
except (json.JSONDecodeError, IndexError, KeyError):
continue
while True:
if await request.is_disconnected():
stream_task.cancel()
break
try:
token = await asyncio.wait_for(token_queue.get(), timeout=0.5)
except asyncio.TimeoutError:
continue
if token is None:
break
full_response += token
yield f"data: {json.dumps({'token': token})}\n\n"
except Exception as e:
from kilostar.utils.logger import get_logger
get_logger("chat_stream").exception(f"Stream error: {e}")
@@ -28,7 +28,7 @@ class ConsciousnessNodeDeps(DepsModel):
"""ConsciousnessNode 在 pydantic-ai Agent 中使用的依赖:原始指令、当前指令以及可用 Skill 列表。"""
original_command: str
command: str
available_skills: Optional[List[str]]
available_skills: Optional[List[str]] = None
class ConsciousnessNodeInput(RequestModel):
"""ConsciousnessNode 各类入参的共同基类,仅用于打 schema 标签。"""
@@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import asyncio
import datetime
from typing import Union
from kilostar.utils.standalone_proxy import actor_class
@@ -125,6 +126,37 @@ class RegulatoryNode:
"""
return await self._run(payload)
async def stream_working(self, payload: MessageRequest, token_queue: "asyncio.Queue") -> None:
"""流式工具调用版本:逐 token 推送到 queue,工具调用结果也会通过 token 输出。
完成后 push None 作为终止信号。
"""
platform = payload.platform
user_name = payload.user_name
message = payload.message
time_str = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
if self.agent is None:
await token_queue.put(None)
return
try:
deps = RegulatoryNodeDeps(
platform=platform,
user_name=user_name,
time=time_str
)
async with self.agent.run_stream(
user_prompt=message, deps=deps, output_type=str
) as stream_result:
async for delta in stream_result.stream_text(delta=True):
await token_queue.put(delta)
except Exception as e:
self.logger.exception(f"RegulatoryNode.stream_working failed: {e}")
await token_queue.put(f"\n\n[错误: {str(e)}]")
finally:
await token_queue.put(None)
async def _run(
self, payload: MessageRequest
) -> Union[MessageResponse, None]: