| name | ag2-ag-ui |
| description | Expose an AG2 `Agent` over the AG-UI protocol so a frontend (CopilotKit, custom React/Next.js, or any AG-UI client) can stream responses (text and reasoning), render tool calls, sync shared state, and track sub-task steps. Wraps the agent with `AGUIStream(agent)` and mounts it in FastAPI via `stream.dispatch(...)` or `stream.build_asgi()`. Use when the user wants a web frontend in front of an AG2 agent rather than a CLI / script. |
| license | Apache-2.0 |
AG-UI integration
When to use
- The user is building a web UI (React / Next.js / anything HTTP+SSE) that should talk to an AG2 agent.
- They want streaming text, reasoning, tool-call rendering, shared state sync, or sub-task step tracking surfaced to the frontend with a standard protocol — not a custom REST/WebSocket contract.
- They're using or considering CopilotKit (the recommended React client).
For a custom narrow-purpose API where you own the contract end-to-end, skip AG-UI and write a plain endpoint instead.
Supported AG-UI features
| Feature | Status |
|---|
Run lifecycle (RUN_STARTED / RUN_FINISHED / RUN_ERROR) | ✓ |
Streaming text events (TEXT_MESSAGE_START / _CONTENT / _END / _CHUNK) | ✓ |
Reasoning events (REASONING_START / _END, REASONING_MESSAGE_START / _CONTENT / _END) | ✓ |
Backend tool lifecycle (TOOL_CALL_START / _ARGS / _RESULT / _END) | ✓ |
Frontend-tool dispatch (TOOL_CALL_CHUNK for client tools in RunAgentInput.tools) | ✓ |
Shared-state snapshots (STATE_SNAPSHOT) | ✓ |
Sub-task steps (STEP_STARTED / STEP_FINISHED, emitted on TaskStarted / TaskCompleted) | ✓ |
| Human input checkpoints surfaced as AG-UI events | ✗ — no human-input event family; hitl_hook passes through to agent.ask |
Installation
pip install "ag2[ag-ui]"
Required. Run this install before delivering the code. If you cannot run commands, state the exact pip install command.
60-second recipe — FastAPI server
from fastapi import FastAPI, Header
from fastapi.responses import StreamingResponse
from ag2 import Agent
from ag2.ag_ui import AGUIStream, RunAgentInput
from ag2.config import OpenAIConfig
agent = Agent(
name="support_bot",
prompt="You help users with billing questions.",
config=OpenAIConfig(model="gpt-4o-mini"),
)
stream = AGUIStream(agent)
app = FastAPI()
@app.post("/chat")
async def run_agent(message: RunAgentInput, accept: str | None = Header(None)) -> StreamingResponse:
return StreamingResponse(
stream.dispatch(message, accept=accept),
media_type=accept or "text/event-stream",
)
uvicorn run_ag_ui:app --reload --port 8000
Test:
curl -N -X POST http://127.0.0.1:8000/chat \
-H "Content-Type: application/json" \
-H "Accept: text/event-stream" \
-d '{"thread_id":"t1","run_id":"r1","messages":[{"id":"m1","role":"user","content":"Hello"}],"state":{},"context":[],"tools":[]}'
Lower-friction alternative — build_asgi()
If you don't need custom auth / middleware, mount the ASGI endpoint directly:
from ag2.ag_ui import AGUIStream
from fastapi import FastAPI
app = FastAPI()
stream = AGUIStream(agent)
app.mount("/chat", stream.build_asgi())
CopilotKit frontend (recommended for React / Next.js)
Two ways to start:
npx copilotkit@latest create -f ag2
The starter layout:
ag2-copilotkit-starter/
├── agent-py/ # Python backend (AG2 + AG-UI)
└── ui-react/ # React + CopilotKit frontend
Backend serves /chat on port 8008 by default. The Next.js route at ui-react/app/api/copilotkit/route.ts bridges the UI to the backend:
import { HttpAgent } from "@ag-ui/client";
import { CopilotRuntime, ExperimentalEmptyAdapter, copilotRuntimeNextJSAppRouterEndpoint } from "@copilotkit/runtime";
import { NextRequest } from "next/server";
const agent = new HttpAgent({ url: "http://localhost:8008/chat" });
const runtime = new CopilotRuntime({ agents: { weather_agent: agent } });
export async function POST(req: NextRequest) {
const { handleRequest } = copilotRuntimeNextJSAppRouterEndpoint({
runtime,
serviceAdapter: new ExperimentalEmptyAdapter(),
endpoint: "/api/copilotkit",
});
return handleRequest(req);
}
Wrap your app:
import { CopilotKit } from "@copilotkit/react-core";
import "@copilotkit/react-ui/styles.css";
export default function RootLayout({ children }: { children: React.ReactNode }) {
return (
<html lang="en">
<body>
<CopilotKit agent="weather_agent" runtimeUrl="/api/copilotkit">
{children}
</CopilotKit>
</body>
</html>
);
}
Render chat (<CopilotChat />), and use useCopilotAction({...}) in client components to render generative UI tied to backend tool calls.
Production checklist
- CORS — allow your frontend origin on the AG-UI backend (
POST, OPTIONS, auth headers).
- Auth — protect both
/api/copilotkit (Next.js route) and backend /chat. Don't rely on client-only secrets.
- SSE buffering — verify backend response is
Content-Type: text/event-stream and that proxy layers (Nginx, CloudFront, etc.) don't buffer SSE.
- Timeouts / retries — conservative values for long-running tool workflows; only retry idempotent requests.
- Tool inputs are untrusted — validate and authorize server-side before invoking privileged tools. Log tool execution with request IDs.
- Tool name parity — frontend
useCopilotAction name must match backend @tool name exactly.
Going deeper
Common pitfalls
- Missing
ag-ui extra — pip install "ag2[ag-ui]" is required; without it from ag2.ag_ui import AGUIStream will fail.
- CORS blocked — frontend can't hit the backend in dev. Add
CORSMiddleware to the FastAPI app for the dev origin.
- No streaming output — proxy or
Content-Type issue. Test with raw curl -N first to isolate.
- Tool UI not rendering — tool name on the React side (
useCopilotAction({ name: "..." })) doesn't exactly match the backend @tool name.
- Putting the endpoint behind a SSE-incompatible proxy — many proxies buffer event-stream responses by default. Disable buffering on the route.
- Trying to use
AGUIStream with classic ConversableAgent — the description here covers ag2.Agent. Classic AG2 agents can be exposed differently; check the AG2 docs for that path.
