Agent Chat UI

Building interactive chat interfaces for LeafMesh agents using the FastAPI services.

API-Based Chat

Use the LeafMesh API server to build chat interfaces:

from leafmesh import LeafMesh
import asyncio

leafmesh = LeafMesh.from_store()

async def chat_loop():
    await leafmesh.start()
    session_id = "chat_session_001"

    print("Chat with the agent (type 'quit' to exit)")
    while True:
        user_input = input("You: ")
        if user_input.lower() == "quit":
            break

        result = await leafmesh.mesh_call(
            "chat_agent",
            input_data={"user_message": user_input},
            session_id=session_id
        )

        response = result.get("response", str(result))
        print(f"Agent: {response}")

    await leafmesh.stop()

asyncio.run(chat_loop())

FastAPI Chat Endpoint

Expose chat functionality as an API:

from fastapi import FastAPI
from pydantic import BaseModel
from leafmesh import LeafMesh

app = FastAPI()
leafmesh = LeafMesh.from_store()

class ChatRequest(BaseModel):
    message: str
    session_id: str

class ChatResponse(BaseModel):
    response: str
    session_id: str
    agent: str

@app.on_event("startup")
async def startup():
    await leafmesh.start()

@app.on_event("shutdown")
async def shutdown():
    await leafmesh.stop()

@app.post("/chat", response_model=ChatResponse)
async def chat(request: ChatRequest):
    result = await leafmesh.mesh_call(
        "chat_agent",
        input_data={"user_message": request.message},
        session_id=request.session_id
    )
    return ChatResponse(
        response=result.get("response", str(result)),
        session_id=request.session_id,
        agent=result.get("_agent", "chat_agent")
    )

Streaming Responses

For real-time chat interfaces, use streaming mode:

agents:
  chat_agent:
    name: "chat_agent"
    model: "gpt-4o"
    stream: true
    prompt: "You are a helpful assistant."
    yields:
      response: "string"

When stream: true, the LLM response is streamed token-by-token. Yields validation runs on the complete response.

Session Management

Each chat session maintains conversation history automatically. The full transcript — role, content, agent attribution, timestamps — is browsable in Studio's Sessions tab, or available via the session REST endpoint (GET /session/{session_id}) for embedding into your own UI.

Multi-Agent Chat

A chat interface that routes to different agents based on the conversation:

agents:
  router:
    name: "router"
    model: "gpt-4o-mini"
    prompt: "Classify the user's request and route to the right specialist."
    yields:
      intent: "string"
    can_call:
      - agent: "tech_support"
        condition: "intent == 'technical'"
      - agent: "billing_support"
        condition: "intent == 'billing'"
      - agent: "general_chat"
        condition: "intent not in ['technical', 'billing']"

The user interacts with a single chat interface, but different agents handle different request types transparently.

Next Steps

Agent Studio — Development environment
FastAPI Integration — Production API setup
Session Management — Session lifecycle

LeafMesh — Interactive chat interfaces