Assistants¶
The gateway doesn't serve OpenAI's Assistants API. Calls to /v1/assistants, /v1/threads (and their runs and
messages) and /v1/vector_stores aren't passed on: they get 404 endpoint_not_found. To have an assistant's calls
governed, build it on an endpoint the gateway serves.
What it is¶
OpenAI's Assistants API keeps assistants, threads, runs and vector stores at OpenAI and runs the model there, out of the gateway's sight. OpenAI itself recommends the Responses API in its place.
What to use instead¶
| You use | Move to | Reaches |
|---|---|---|
| A thread and a run per conversation | Responses, with previous_response_id |
OpenAI and Azure OpenAI |
| Function tools, your own conversation history | Chat Completions with tools |
Every provider: GPT, Claude, Gemini |
| An agent that plans and calls tools | An agent framework on Chat Completions, such as LangChain | Every provider |
| File search over a vector store | Your own retrieval: Embeddings, then Chat Completions | Every provider for the answer |
A run that was a single model turn maps directly onto client.chat.completions.create(...).
Check it¶
curl -i https://<your-gateway>/v1/assistants -H "Authorization: Bearer $OPENAI_API_KEY"
A 404 with x-vidai-error-code: endpoint_not_found confirms the path isn't served.
Good to know¶
- Calls an app makes straight to OpenAI's Assistants API, around the gateway, don't show in Request logs or Cost.