Real-Time Streaming¶

When a prompt is sent, the agent runs on a Celery worker and publishes events to Redis. The API's SSE endpoint relays these events to any connected client.

Architecture¶

sequenceDiagram
    participant Client
    participant API as FastAPI API
    participant Redis
    participant Worker as Celery Worker
    participant SDK as Copilot SDK

    Client->>API: POST /prompt
    API->>Redis: Enqueue task
    API-->>Client: 201 Accepted
    Client->>API: GET /stream (SSE)
    Redis->>Worker: Deliver task
    Worker->>SDK: session.send(prompt)
    loop Agent loop
        SDK->>Worker: Events (tokens, tool calls, etc.)
        Worker->>Redis: Publish to channel
        Redis->>API: Deliver event
        API-->>Client: SSE event
    end
    Worker->>Redis: Publish status: completed
    Redis->>API: Deliver final event
    API-->>Client: SSE: status completed

Event Types¶

Event	Payload	Description
`log`	`{event, detail}`	Agent lifecycle: tool calls, errors, compaction
`message`	`{role, content}`	Complete assistant or tool message
`message_delta`	`{delta}`	Streaming token — live character-by-character rendering
`usage`	`{total_in, total_out, cost...}`	Cumulative token counts, premium requests, and cost
`status`	`{status, current_turn}`	Workflow state changes (running → completed/failed)

Connecting¶

curlJavaScriptPython

curl -N http://localhost:8000/api/workflows/<WF_ID>/stream

const es = new EventSource('/api/workflows/<WF_ID>/stream');
es.onmessage = (e) => {
  const { type, data } = JSON.parse(e.data);
  // type: log | message | message_delta | usage | status
};

import requests

with requests.get(
    f"http://localhost:8000/api/workflows/{wf_id}/stream",
    stream=True
) as resp:
    for line in resp.iter_lines():
        if line:
            print(line.decode())

Event Bus¶

Events are published to Redis channel workflow:events:{workflow_id}:

{
  "type": "log",
  "data": {"event": "tool_call", "detail": "Called search_issues"},
  "timestamp": "2026-04-10T12:00:00+00:00"
}

The built-in dashboard uses this SSE stream to render live logs, token-by-token message streaming, and real-time usage metrics.