Streaming

Open an SSE stream

Enable stream: true and disable client buffering. The server sends data events as response parts become available.

Process the stream incrementally: one event does not have to contain completed text or a complete tool call.

curl --no-buffer --request POST https://cicora.ai/api/v1/chat/completions \
  --header "Authorization: Bearer $CICORA_API_KEY" \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "openai/gpt-5.6-sol",
    "stream": true,
    "stream_options": {"include_usage": true},
    "messages": [{"role": "user", "content": "Write a three-line poem."}]
  }'

Assemble deltas

• Accumulate text by choice/index or item identifier.
• Collect tool arguments as a stream of JSON fragments and parse them after completion.
• When final metrics matter, explicitly enable the available usage option in the stream.

Finish and cancel

Treat the stream as complete only after a terminal event. A connection closing without one is a distinct failure path.

Attach an AbortController or equivalent to user navigation so an unused request does not remain open.