Spaces:
Paused
Paused
File size: 2,426 Bytes
a348131 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 | from langgraph.graph import StateGraph, END
from app.agent.state import AgentState
from langchain_core.messages import HumanMessage, AIMessage, SystemMessage
from app.services.mcp_client import get_all_tools, mcp_clients
import json
# Placeholder for RAG
async def retrieve(state: AgentState):
# Simulating RAG retrieval
query = state['messages'][-1].content
# In reality, query vector DB
context = f"Retrieved context for: {query}"
return {"context": context}
# Placeholder for LLM / Think
async def think(state: AgentState):
messages = state['messages']
context = state.get('context', '')
# Simple logic to simulate "thinking"
last_msg = messages[-1].content.lower()
# If user asks to "search", we trigger tool call
if "search" in last_msg:
# Mocking a tool call decision
return {"current_step": "tool_call", "tool_calls": [{"name": "brave-search", "args": {"q": "search term"}}]}
return {"current_step": "generate"}
# Tool execution node
async def tool_call_node(state: AgentState):
# This would parse the decision from 'think'
# For now, we just list tools or call a mock one
# Assuming 'think' decided to call a tool
# Example: fetch all tools
tools = await get_all_tools()
# Mock execution
# result = await client.call_tool(...)
return {"messages": [AIMessage(content=f"Executed tool. Available tools: {len(tools)}")]}
# Generation node
async def generate(state: AgentState):
context = state.get('context', '')
return {"messages": [AIMessage(content=f"Generated response based on: {context}")]}
def route_step(state: AgentState):
step = state.get('current_step')
if step == "tool_call":
return "tool_call"
return "generate"
# Build Graph
builder = StateGraph(AgentState)
builder.add_node("retrieve", retrieve)
builder.add_node("think", think)
builder.add_node("tool_call", tool_call_node)
builder.add_node("generate", generate)
builder.set_entry_point("retrieve")
builder.add_edge("retrieve", "think")
builder.add_conditional_edges("think", route_step, {
"tool_call": "tool_call",
"generate": "generate"
})
builder.add_edge("tool_call", "generate") # Loop back or go to generate? User said "Tool Call -> Generate"
builder.add_edge("generate", END)
graph = builder.compile()
|