Real-time & WebSockets

Request/response is a letter; realtime is a phone call. WebSockets give your FastAPI apps a live wire — streaming LLM tokens, chat, dashboards — and webhooks give the outside world a way to call YOU.

What you'll learn

1. WebSockets: the live wire
2. Building realtime apps
3. Webhooks: inbound events
4. Production hardening

Remember this

• WebSocket upgrades HTTP into a full-duplex frame pipe — either side pushes anytime; delivery isn't guaranteed, so acks are your protocol's job
• Connection manager (rooms/broadcast) + token streaming are the two patterns — async lets one thread interleave thousands of live connections
• Webhooks invert HTTP: verify HMAC signatures on raw bodies, dedupe by event ID, ACK in milliseconds and process via queue
• Production needs heartbeats (zombie detection), bounded queues (backpressure), and a pub/sub backbone (Redis) to scale past one replica

Code: Realtime, end to end

# The live pipeline:
#   browser ↔ WebSocket ↔ FastAPI ↔ LLM stream
#   world   → webhook  → FastAPI → queue → broadcast

@app.websocket("/live/{room}")
async def live(ws: WebSocket, room: str):
    await manager.join(room, ws)
    beat = asyncio.create_task(heartbeat(ws))
    try:
        async for token in llm.stream(await ws.receive_text()):
            await ws.send_text(token)          # tokens as they're born
        await ws.send_text("[DONE]")
    except WebSocketDisconnect:
        pass
    finally:
        beat.cancel()

# Heartbeat in the background, tokens in the foreground,
# disconnects handled in the finally — a socket that
# survives production is a socket with all three.