OpenAI-compatible services
OpenAI-compatible is any cloud service that speaks the OpenAI chat interface: a gateway in front of several models, or a regional provider, for example where OpenAI, Anthropic and Google are not available. You need the service's own address and its own key, and the service bills you. A server on your own computers is set up as Your own server instead: see vLLM and other compatible servers.
How to set it up
- Settings ▸ Assistant ▸ OpenAI-compatible ("Another cloud service at its own address. Needs its key.").
- Service address: its OpenAI-compatible address in full, with https, usually ending in
/v1. The service names it in its documentation. - Paste the service's own key. The key field waits for the address: "Type the service address above first: the key belongs to that address." The key goes only to that address.
- Model: pick it from the service's list, or type its name. Choose a model that can call tools.
- Test connection, then Save settings.
Remember on this computer keeps the key in the system keychain, as for the other cloud providers. See Keys and where they are kept.
Rules
- The address must start with https: "The service address is needed in full, with https, for example https://api.example.com/v1." A cloud key is never sent to an http address.
- A key belongs to its address. For another address, paste its own key.
- The app shows no prices for such a service. See the service's own page.
- If your network needs a proxy, set it under Settings ▸ Assistant ▸ Network ▸ Proxy (optional).
What can go wrong
- "A compatible service needs its https address. Type it under Settings ▸ Assistant."
- "The server did not accept the key. Check the key in the assistant settings."
- "The server does not have this model, or it was retired. Choose another model in the assistant settings."
- "The server answered, but not as a model. Check the address in the assistant settings."
See Failed answers.