Streaming and WebSocket

Codex Relay supports regular HTTP responses, SSE streaming, and a Responses WebSocket endpoint for compatible clients.

SSE streaming

Set this field in the request body:

{
  "stream": true
}

Use -N with curl to disable local buffering:

curl -N https://codexapi.asia/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MODEL_ID",
    "input": "Count from one to five.",
    "stream": true
  }'

Streaming displays initial output sooner, but it does not reduce the model's actual computation time.

WebSocket

The Responses WebSocket URL is:

wss://codexapi.asia/v1/responses

Enable it only when the client explicitly supports Responses WebSocket. Standard OpenAI SDK calls and Chat Completions clients generally use HTTP/SSE.

Proxy considerations

• Cloudflare and reverse proxies must allow long-lived connections and WebSocket Upgrade.
• Client timeouts should include queueing, inference, and tool execution time.
• Never cache /v1/* API responses.
• After a disconnect, retry only requests that are idempotent or safe to replay.

Choose a transport

Mode	Best for
Non-streaming HTTP	Scripts, short requests, one complete JSON object
SSE	Chat, Codex, OpenAI SDK apps, and long text generation
WebSocket	Clients that explicitly implement Responses WS and persistent sessions