Stream tokens as they are generated

Enable streaming

Set stream: true on a chat completions request. The response is text/event-stream with OpenAI-style chunks, terminated by a data: [DONE] sentinel.

curl -N https://api.hardcarrx.com/v1/chat/completions \
  -H "Authorization: Bearer $HARDCARRX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "openai",
    "model": "gpt-4.1-mini",
    "stream": true,
    "messages": [{"role": "user", "content": "Count to five slowly"}]
  }'

Each data event carries a chat.completion.chunk payload:

data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"2"},"finish_reason":null}]}

data: [DONE]

Cancellation

Closing the connection cancels the exchange. Cancellation propagates upstream: the provider poll is stopped and no further tokens are billed into memory writeback. A slow client pauses upstream polling instead of accumulating unbounded memory.

Bounded resources

• A single SSE line or event larger than the platform's bounded limits aborts the stream rather than buffering unboundedly.
• Invalid upstream events are handled explicitly; raw upstream data payloads are forwarded verbatim.

Responses endpoint

The /v1/responses compatibility layer is non-streaming in its first phase. Sending stream: true is rejected explicitly with 422 — it never silently degrades to a non-streaming call.

curl https://api.hardcarrx.com/v1/responses \
  -H "Authorization: Bearer $HARDCARRX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1-mini",
    "stream": true,
    "input": "Hello"
  }'
# 422 — streaming is not yet supported by the /v1/responses compatibility layer