Referensi API

Streaming

OpenAI-style

Set "stream": true. Respons bertipe text/event-stream berisi chunk: `` data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"delta":{"content":"Ha"}}]} data: {"choices":[{"delta":{},"finish_reason":"stop"}],"usage":{"prompt_tokens":17,"completion_tokens":11,"total_tokens":28}} data: [DONE] Tambahkan stream_options: {"include_usage": true}` agar chunk terakhir memuat usage.

Anthropic-style

Event: message_start → content_block_start → content_block_delta (bertipe text_delta / thinking_delta) → content_block_stop → message_delta (usage final) → message_stop.

Perilaku penting

  • Streaming PixzRouter adalah true streaming — chunk diteruskan begitu tiba dari model, tanpa buffer penuh.
  • Stream terputus di tengah? Anda hanya ditagih token yang tercatat pada usage — tanpa data usage, tidak ada tagihan.
  • Header response X-Pixz-Request-Id selalu tersedia untuk audit.

Batalkan stream

Putuskan koneksi (abort). Upstream request ikut dibatalkan; penagihan berhenti pada usage tercatat.