OpenAI API, endpoint by endpoint¶
Code written for the OpenAI API keeps working when you point it at the gateway. This section takes each endpoint in turn: what reaches which provider, what the gateway records, and the errors you can meet.
The settings are the same for every endpoint: the base address https://<your-gateway>/v1 and a Vidai key. See
OpenAI SDK for Python for how to set them.
Which endpoint reaches which provider¶
Chat Completions is translated: one request shape reaches every provider. Every other endpoint goes only to a provider that speaks the OpenAI API itself: OpenAI, Azure OpenAI, and providers added with the OpenAI kind of API (Mistral, Groq, Together AI and others) where that provider offers the endpoint. Sent to a Claude or Gemini model, such a call gets a clear 400 and is never sent somewhere else.
| Endpoint | Path | Which providers | Page |
|---|---|---|---|
| Chat Completions | POST /v1/chat/completions |
All: OpenAI, Azure OpenAI, Anthropic, Gemini, Vertex AI, AWS Bedrock and OpenAI-compatible providers | Chat Completions |
| Responses | POST /v1/responses and /v1/responses/{id} |
OpenAI and Azure OpenAI (and OpenAI-compatible providers that offer it) | Responses |
| Completions (legacy) | POST /v1/completions |
OpenAI and Azure OpenAI (and OpenAI-compatible providers that offer it) | Completions |
| Embeddings | POST /v1/embeddings |
OpenAI-format providers | Embeddings |
| Moderations | POST /v1/moderations |
OpenAI and Azure OpenAI | Moderations |
| Images | POST /v1/images/generations, /edits, /variations |
OpenAI-format providers | Images |
| Audio | POST /v1/audio/transcriptions, /translations, /speech |
OpenAI-format providers | Audio |
| Video | POST /v1/video/generations, /v1/videos |
OpenAI-format providers that still offer video (OpenAI itself retired it) | Video |
| OCR, rerank, predictions | POST /v1/ocr, /v1/rerank, /v1/predictions |
Providers that offer them | OCR, rerank and predictions |
| Files and uploads | /v1/files, /v1/uploads |
The one provider that serves files | Files and uploads |
| Batches | /v1/batches |
The one provider that serves files | Batches |
| Fine-tuning | /v1/fine_tuning/jobs |
The one provider that serves files | Fine-tuning |
| Models | GET /v1/models |
Answered by the gateway: the models your key may call | Chat Completions |
| Assistants | /v1/assistants, /v1/threads, /v1/vector_stores |
Not served | Assistants |
| Realtime | /v1/realtime (WebSocket) |
Not served | Realtime |
The same on every endpoint¶
- Every answer carries
x-vidai-request-id. Paste it into Request logs › Request reference to find the call (it shows within seconds). - Errors carry
x-vidai-error-code. Match on it, not on the message. The full list is in Headers and errors. - Paths are exact. A near miss, such as
/v1/audio/transcriptionfor/v1/audio/transcriptions, gets 404endpoint_not_found. - Cost is never in the answer. Find it in Request logs.