Stream responses - Aixy

After connecting a provider and creating a key, set stream to true to receive incremental server-sent events. Replace the example model with one available to your project:

curl --no-buffer https://api.aixy-gateway.com/v1/chat/completions \
  --header "Authorization: Bearer $AIXY_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "openai/gpt-4.1-mini",
    "stream": true,
    "max_completion_tokens": 256,
    "messages": [
      {"role": "user", "content": "Write a short launch announcement."}
    ]
  }'

On matching native routes, Aixy forwards the provider’s event stream without reconstruction. On translated routes, it emits incremental events in the requested client format. Choose a model that supports streaming and test the protocol your application uses.

Client behavior

Set a connection timeout separately from the total streaming timeout.
Process complete SSE events rather than assuming each network chunk is one event.
Treat provider error events and non-2xx HTTP responses as distinct failure paths.
Stop reading and close the connection when the end user cancels generation.
Record your X-Request-ID so an individual stream can be correlated with gateway telemetry.

Usage can arrive near the end of a provider stream. Do not estimate final billing solely from partial output observed by the client. Review pricing and cost attribution for final-usage settings, partial accounting, and hard-budget settlement.