Streaming responses

Set stream to true to receive content as the model generates it instead of waiting for the complete answer.

cURL

Use --no-buffer so events are printed as they arrive:

curl --no-buffer --silent --show-error \
  'https://omrouter.com/v1/chat/completions' \
  --header "Authorization: Bearer $OMROUTER_API_KEY" \
  --header 'Content-Type: application/json' \
  --data "{
    \"model\": \"$OMROUTER_MODEL_ID\",
    \"stream\": true,
    \"messages\": [
      {\"role\": \"user\", \"content\": \"Count from one to five.\"}
    ]
  }"

Chat Completions streams contain data: events. A normal stream ends with the protocol’s terminal event, commonly data: [DONE].

Python SDK

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["OMROUTER_API_KEY"],
    base_url="https://omrouter.com/v1",
)

stream = client.chat.completions.create(
    model=os.environ["OMROUTER_MODEL_ID"],
    stream=True,
    messages=[{"role": "user", "content": "Count from one to five."}],
)

for chunk in stream:
    text = chunk.choices[0].delta.content
    if text:
        print(text, end="", flush=True)

JavaScript SDK

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.OMROUTER_API_KEY,
  baseURL: 'https://omrouter.com/v1',
});

const stream = await client.chat.completions.create({
  model: process.env.OMROUTER_MODEL_ID,
  stream: true,
  messages: [{ role: 'user', content: 'Count from one to five.' }],
});

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

Handle incomplete streams

An HTTP 200 only means the connection has started. The answer is complete after the SDK finishes normally or the stream sends its end marker.

• Show an interrupted state when the connection ends early.
• Do not retry automatically after displaying text or running a tool; that can duplicate content or actions.
• Apply both connection and total-duration timeouts.
• Stop reading when the client request is cancelled.
• Do not log API keys, complete prompts, or model answers.

Chat Completions, Responses, Anthropic, and Gemini use different streaming formats. Use the matching SDK reader for each format.