ReferenceReference

API reference

Every route, method, and body format the gateway accepts.

By the end: You know which endpoint to call for each model family.

Base URL: http://localhost:3000. Bodies follow each family's native format and are forwarded without translation, so use the SDK that matches the model.

Method	Path	Purpose
GET	/v1/models	List available models (OpenAI list format).
POST	/v1/chat/completions	OpenAI Chat Completions. Streaming supported.
POST	/v1/responses	OpenAI Responses. Streaming supported.
POST	/v1/messages	Anthropic Messages. Streaming supported.
POST	/v1/messages/count_tokens	Anthropic token counting. Not billed.
POST	/v1beta/models/{model}:generateContent	Gemini generateContent.
POST	/v1beta/models/{model}:streamGenerateContent	Gemini streaming. Add ?alt=sse.
POST	/v1/images/generations	Text-to-image. Billed per image.
POST	/v1/videos/generations	Text-to-video job. Billed on completion.

Request rules

model is required — in the body, or in the path for Gemini routes — and must be returned by GET /v1/models.
Bodies must be JSON objects and are limited to 1 MiB by default.
Redirects from the upstream service are never followed.
Your credentials are never forwarded; only anthropic-version and anthropic-beta pass through.

See streaming and limits and timeouts.