aboutme_chat/langgraph_service/supervisor_agent.py

from typing import TypedDict, Annotated, Sequence
from langgraph.graph import StateGraph, END
from langchain_core.messages import BaseMessage, HumanMessage, AIMessage
import operator
import httpx
import os
import logging

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# State definition
class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], operator.add]
    next_agent: str
    context: dict

# Agent routing logic
def supervisor_node(state: AgentState):
    """Supervisor decides which specialist agent to call."""
    last_message = state["messages"][-1].content.lower()
    
    # Simple routing logic based on keywords
    if any(kw in last_message for kw in ["repo", "code", "git", "github", "gitea", "project", "development"]):
        return {"next_agent": "librarian"}
    elif any(kw in last_message for kw in ["write", "edit", "create", "fix", "bug", "implement", "code change"]):
        return {"next_agent": "opencode"}
    elif any(kw in last_message for kw in ["sam", "hobby", "music", "experience", "skill", "about"]):
        return {"next_agent": "librarian"}
    else:
        return {"next_agent": "brain"}  # Default to general LLM

def librarian_agent(state: AgentState):
    """Librarian agent - queries knowledge base (ChromaDB)."""
    last_message = state["messages"][-1].content
    
    try:
        # Call knowledge service
        response = httpx.post(
            "http://knowledge-service:8080/query",
            json={"question": last_message},
            timeout=10.0
        )
        
        if response.status_code == 200:
            context = response.json().get("context", "")
            return {
                "messages": [AIMessage(content=f"Based on my knowledge base:\n\n{context}")],
                "context": {"source": "librarian", "context": context}
            }
    except Exception as e:
        logger.error(f"Librarian error: {e}")
    
    return {
        "messages": [AIMessage(content="I couldn't find relevant information in the knowledge base.")],
        "context": {"source": "librarian", "error": str(e)}
    }

def opencode_agent(state: AgentState):
    """Opencode agent - handles coding tasks via MCP."""
    last_message = state["messages"][-1].content
    
    # Placeholder - would integrate with opencode-brain
    return {
        "messages": [AIMessage(content=f"I'm the coding agent. I would help you with: {last_message}")],
        "context": {"source": "opencode", "action": "coding_task"}
    }

def brain_agent(state: AgentState):
    """Brain agent - general LLM fallback."""
    last_message = state["messages"][-1].content
    
    try:
        # Call opencode-brain service
        auth = httpx.BasicAuth("opencode", os.getenv("OPENCODE_PASSWORD", "sam4jo"))
        timeout_long = httpx.Timeout(180.0, connect=10.0)
        
        with httpx.AsyncClient(auth=auth, timeout=timeout_long) as client:
            # Create session
            session_res = client.post("http://opencode-brain:5000/session", json={"title": "Supervisor Query"})
            session_id = session_res.json()["id"]
            
            # Send message
            response = client.post(
                f"http://opencode-brain:5000/session/{session_id}/message",
                json={"parts": [{"type": "text", "text": last_message}]}
            )
            
            data = response.json()
            if "parts" in data:
                for part in data["parts"]:
                    if part.get("type") == "text":
                        return {
                            "messages": [AIMessage(content=part["text"])],
                            "context": {"source": "brain"}
                        }
    except Exception as e:
        logger.error(f"Brain error: {e}")
    
    return {
        "messages": [AIMessage(content="I'm thinking about this...")],
        "context": {"source": "brain"}
    }

def route_decision(state: AgentState):
    """Routing function based on supervisor decision."""
    return state["next_agent"]

# Build the graph
workflow = StateGraph(AgentState)

# Add nodes
workflow.add_node("supervisor", supervisor_node)
workflow.add_node("librarian", librarian_agent)
workflow.add_node("opencode", opencode_agent)
workflow.add_node("brain", brain_agent)

# Add edges
workflow.set_entry_point("supervisor")

# Conditional routing from supervisor
workflow.add_conditional_edges(
    "supervisor",
    route_decision,
    {
        "librarian": "librarian",
        "opencode": "opencode",
        "brain": "brain"
    }
)

# All specialist agents end
workflow.add_edge("librarian", END)
workflow.add_edge("opencode", END)
workflow.add_edge("brain", END)

# Compile the graph
supervisor_graph = workflow.compile()

# Main entry point for queries
async def process_query(query: str) -> dict:
    """Process a query through the supervisor graph."""
    result = await supervisor_graph.ainvoke({
        "messages": [HumanMessage(content=query)],
        "next_agent": "",
        "context": {}
    })
    
    return {
        "response": result["messages"][-1].content,
        "context": result.get("context", {})
    }