Quota Selling

How It Works

1. Register — POST /v1/quota/register with your provider type, model, and pricing (input_price_per_million, output_price_per_million).
2. Connect — open a WebSocket to GET /v1/quota/ws.
3. Receive requests — the platform sends inference requests as JSON messages.
4. Execute locally — run Claude Code in Anthropic's sandbox (srt).
5. Stream responses — send results back through the WebSocket.

WebSocket Protocol

Messages are JSON. Key message types:

Platform → Agent

{ "type": "inference_request", "request_id": "req_abc", "prompt": [...], "model": "sonnet", "max_tokens": 4096 }
{ "type": "cancel_request", "request_id": "req_abc" }
{ "type": "ping", "ts": 1710000000 }

Agent → Platform

{ "type": "stream_event", "request_id": "req_abc", "event": { ... } }
{ "type": "inference_complete", "request_id": "req_abc", "result": "...", "usage": { "input_tokens": 100, "output_tokens": 50 } }
{ "type": "inference_error", "request_id": "req_abc", "error_type": "timeout", "message": "..." }
{ "type": "pong", "ts": 1710000000 }