API overview

Choose a protocol first, then use the key-specific model catalog and the complete API reference.

Task	Route family	First step
General chat	OpenAI Chat	POST /v1/chat/completions
Responses and coding tools	OpenAI Responses	POST /v1/responses
Claude-compatible clients	Anthropic Messages	POST /v1/messages
Gemini clients	Gemini actions	POST /v1beta/models/{model}:generateContent
Vectors, ranking or moderation	dedicated JSON routes	GET /v1/models, then select a model
Images or audio	media routes	choose JSON or multipart as documented
Live sessions	Realtime WebSocket	GET /v1/realtime?model=MODEL

Every request uses a key issued in Telegram and is filtered by that key's endpoint and model policy. Read GET /v1/models before hard-coding a model, send an explicit output limit, and keep the returned x-request-id with any support report.

The authentication guide explains headers and base URLs; supported endpoints is the quick inventory.