Streaming

Add "stream": true. The client must read complete SSE events separated by a blank line.

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"delta":{"content":"Hel"}}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"delta":{"content":"lo"}}]}

data: [DONE]

Usage normally arrives in terminal events. Do not finalize local accounting before processing the terminal event.

• Close the upstream reader when the client cancels.
• Retry only requests for which the client received no useful bytes.
• Keep X-GYW-Request-ID for diagnostics.