40 lines
1.5 KiB
Python
40 lines
1.5 KiB
Python
from langchain_core.messages import HumanMessage
|
|
|
|
async def run_agent_with_streaming(app,query:str):
|
|
"""
|
|
通用流式运行器，负责将 LangGraph 的运行过程可视化输出到控制台
|
|
|
|
:param app: 编译好的 LangGraph 应用 (workflow.compile())
|
|
:param query: 用户输入的问题
|
|
"""
|
|
print(f'\n用户:{query}\n')
|
|
print("🤖 AI:",end="",flush=True)
|
|
|
|
# 构造输入消息
|
|
inputs = {"messages":[HumanMessage(content=query)]}
|
|
|
|
# 核心:监听v2版本的事件流(相比v1更全面)
|
|
async for event in app.astream_events(inputs,version="v2"):
|
|
kind = event["event"]
|
|
|
|
# 1.监听LLM的流式吐字(嘴在动)
|
|
if kind == "on_chat_model_stream":
|
|
chunk = event["data"]["chunk"]
|
|
# 过滤掉空的chunk(有时工具调用会产生空内容)
|
|
if chunk.content:
|
|
print(chunk.content,end="",flush=True)
|
|
|
|
# 2.监听工具开始调用(手在动)
|
|
elif kind == "on_tool_start":
|
|
tool_name = event["name"]
|
|
# 不打印内部包装，只打印自定义的工具
|
|
if not tool_name.startswith("_"):
|
|
print(f"\n\n🔨 正在调用工具: {tool_name} ...")
|
|
|
|
# 3.监听工具调用结束(拿到结果)
|
|
elif kind == "on_tool_end":
|
|
tool_name = event["name"]
|
|
if not tool_name.startswith("_"):
|
|
print(f"✅ 调用完成，继续思考...\n")
|
|
print("🤖 AI: ", end="", flush=True)
|
|
print("\n\n😊 输出结束!") |