Streaming responses

Set stream: true in a chat-completion request to receive incremental output. The HTTP connection remains open while the server sends events.

Event format

The response content type is text/event-stream. A typical text-delta event looks like this illustrative example:

data: {"choices":[{"index":0,"delta":{"content":"Hello"}}]}

data: [DONE]

Do not assume every chunk contains text. SDKs may expose empty deltas, finish reasons, usage data or model-specific fields. Process the fields your application understands and preserve the ability to handle future fields.

JavaScript client

Use the official OpenAI JavaScript client on your server to handle SSE framing rather than splitting raw network packets yourself.

npm install openai

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.STREAMINGTELLIGENCE_API_KEY,
  baseURL: process.env.STREAMINGTELLIGENCE_BASE_URL,
});

const stream = await client.chat.completions.create({
  model: process.env.STREAMINGTELLIGENCE_MODEL,
  messages: [{ role: 'user', content: 'Describe this moment.' }],
  stream: true,
});

try {
  for await (const chunk of stream) {
    process.stdout.write(chunk.choices[0]?.delta?.content ?? '');
  }
} finally {
  stream.controller.abort();
}

Disconnections and retries

Cancel the request when the user leaves or no longer needs the answer. Treat a connection that closes before completion as interrupted; do not silently concatenate a second independent generation onto the first.

Retry transient failures with bounded exponential backoff. Automatic retries can consume additional usage. Authentication and request-validation errors need a corrected request, not repeated retries.

Multimodal streaming

Supported input formats, context limits and session transports depend on the selected model. Follow its published interface contract when integrating continuous camera input or historical context; do not assume a chat-completion endpoint also accepts WebSocket connections.