Streaming

Streaming allows you to receive model output in real time as it’s being generated, improving user experience and perceived response speed.

How It Works

kineticRouter uses the Server-Sent Events (SSE) protocol for streaming:

The client sends a request with stream: true
The server progressively returns generated content chunks
Each chunk is sent via SSE with a data: prefix
data: [DONE] is sent when generation is complete

OpenAI Protocol Streaming

cURL

Terminal


curl https://api.kineticrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $KINETICROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -N \
  -d '{
    "model": "openai/gpt-4o",
    "messages": [{"role": "user", "content": "Write a poem about programming"}],
    "stream": true
  }'

Python

stream_openai.py


from openai import OpenAI
 
client = OpenAI(
    base_url="https://api.kineticrouter.com/v1",
    api_key="<your KINETICROUTER_API_KEY>"
)
 
stream = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{"role": "user", "content": "Write a poem about programming"}],
    stream=True
)
 
for chunk in stream:
    content = chunk.choices[0].delta.content
    if content:
        print(content, end="", flush=True)

TypeScript

stream_openai.ts


import OpenAI from 'openai'
 
const client = new OpenAI({
  baseURL: 'https://api.kineticrouter.com/v1',
  apiKey: '<your KINETICROUTER_API_KEY>'
})
 
const stream = await client.chat.completions.create({
  model: 'openai/gpt-4o',
  messages: [{ role: 'user', content: 'Write a poem about programming' }],
  stream: true
})
 
for await (const chunk of stream) {
  const content = chunk.choices[0]?.delta?.content
  if (content) process.stdout.write(content)
}

Anthropic Protocol Streaming

Python

stream_anthropic.py

Planned example

The Anthropic-native route is planned and its configuration is shown only as a reference. This example is intentionally not copyable in its current display state.

Streaming + Function Calling

Streaming also supports Function Calling scenarios. The model will stream the tool call request, and you can continue the conversation after processing:

stream_with_tools.py


stream = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{"role": "user", "content": "What's the weather like in San Francisco today?"}],
    tools=[{
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get weather for a given city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "City name"}
                },
                "required": ["city"]
            }
        }
    }],
    stream=True
)
 
for chunk in stream:
    delta = chunk.choices[0].delta
    if delta.tool_calls:
        # Handle tool call
        print(f"Calling tool: {delta.tool_calls[0].function}")
    elif delta.content:
        print(delta.content, end="", flush=True)

Error Handling and Reconnection

Streaming connections may be interrupted by network issues. We recommend implementing reconnection logic.

stream_retry.py


import time
 
def stream_with_retry(client, max_retries=3, **kwargs):
    for attempt in range(max_retries):
        try:
            stream = client.chat.completions.create(stream=True, **kwargs)
            for chunk in stream:
                yield chunk
            return  # Completed successfully
        except Exception as e:
            if attempt < max_retries - 1:
                wait = 2 ** attempt  # Exponential backoff
                print(f"\nConnection interrupted, retrying in {wait}s...")
                time.sleep(wait)
            else:
                raise e

Best Practices

Always set timeouts — Avoid indefinite waiting
Handle incomplete chunks — Some chunks may not contain content
Implement reconnection — Use exponential backoff strategy
Use flush on the frontend — Ensure content is displayed immediately