Stream a chat response

Set stream: true to receive the response incrementally. Keep the HTTP connection open, parse complete SSE events and append each content delta once.

curl --no-buffer --fail-with-body https://api.shareai.now/api/v1/chat/completions \
  -H "Authorization: Bearer $SHAREAI_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{"model":"MODEL_IDENTIFIER","messages":[{"role":"user","content":"Explain server-sent events briefly."}],"stream":true}'

Read event boundaries#

SSE separates events with a blank line. Network chunks can split a JSON value or contain multiple events, so do not parse each network read as a complete JSON response. Decode UTF-8 incrementally and retain incomplete data until the event is complete.

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

data: {"object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

Render content once#

Append choices[0].delta.content when present. Preserve whitespace and repeated words. A role-only or final chunk can have no content. Treat [DONE] as the stream terminator.

Handle stream errors#

Check the HTTP status before entering the SSE parser. After streaming starts, handle an event: error event and unexpected disconnection. Mark an interrupted answer as incomplete and let the user retry explicitly. Do not append a second independently generated answer to partial output.

Cancel from your application#

Use your HTTP client’s cancellation mechanism when the user stops generation. Keep cancellation, a network failure and a completed answer as separate UI states. Do not promise that closing a client connection reverses usage already processed.