Requesty (direct)

Direct calls to the Requesty router's OpenAI-compatible API (base_url=https://router.requesty.ai/v1), instrumented via the OpenAI SDK wrapper — no first-party Requesty wrapper exists in telara-observability today, but because Requesty is OpenAI-compatible by design ("One API. 300+ models."), the generic OpenAI wrapper pointed at the router's base URL captures spans the same way it does for any OpenAI-compatible endpoint. pip extra: pip install telara-observability[openai] Requesty returns a `cost` field inline in the `usage` object on every non-streaming response — a cost source most routers do not expose — so cost_paths checks for it before falling back to model_price_table derivation. Requesty sits between the customer and the underlying model vendor (OpenAI, Anthropic, etc.), so cost attribution here lands on Requesty as the router, not on the origin model provider — a real chargeback limitation: the origin vendor's own admin-API pull (if connected) will not show these calls.