Endpoints at a glance
Before you point a tool at the gateway, check that the calls it makes are ones the gateway carries. Chat reaches every
provider; everything else reaches a provider of the same format; a few APIs aren't carried at all. The gateway never
quietly sends a call somewhere it can't be served: it answers with a clear error instead.
Translated across providers
| Path |
Format |
Reaches |
POST /v1/chat/completions |
OpenAI |
OpenAI, Azure OpenAI, Anthropic, Gemini, Vertex AI, Amazon Bedrock |
POST /v1/messages |
Anthropic |
the same |
POST /v1beta/models/{model}:generateContent, :streamGenerateContent |
Gemini |
the same |
Vertex AI's own …/publishers/{google\|anthropic}/models/{model}:generateContent |
Gemini (Vertex) |
the same |
A model of another kind of provider gets a 400 (bad_request) that names the problem.
Answered by the gateway itself
| Path |
What you get |
GET /v1/models, GET /v1beta/models |
The models your key may call |
GET /health |
{"status":"ok"}, no key needed |
Not carried
| API |
Page |
OpenAI Assistants (/v1/assistants, /v1/threads, /v1/vector_stores) |
Assistants API |
OpenAI Realtime (/v1/realtime, WebSocket) |
Realtime API |
Anthropic message batches (/v1/messages/batches) |
Message batches |
Any other path gets a 404 with x-vidai-error-code: endpoint_not_found; a path the gateway carries, with a model it
doesn't know, gets 404 model_not_found.
Good to know
- Files, batches and uploads name no model, so the gateway sends them to the one provider that serves files, or the
one you name with the
x-provider header.
- The full reference, with methods and every detail: Endpoints.