OpenAI-compatible endpoint

Bring the business's own OpenAI-compatible endpoint key.

What it is

Any endpoint that speaks the OpenAI API — a self-hosted model, a regional provider, or a gateway a client already runs. This is the escape hatch for everything not listed above.

When to reach for it

Use it when a client's model must stay on their own infrastructure, which is a real requirement for some financial and health data. Their model, their server, RAiN only sends the request.

What you need

An API key from OpenAI-compatible endpoint, belonging to the business this project is for. It is encrypted at rest and only ever used to run that project's agents.

It also needs the endpoint's base URL, which must be https.

Connecting it

Model keys belong to one project, like every other credential in RAiN AI Studio. A package delivered to a business runs on that business's key and bills that business — not the expert who built it, and not RAiN.

Open the project and stay on the Visual Editor tab.
Open an agent and choose one of these models.
If the project has no key for it yet, the form asks for one there and then.

The workspace-level LLM Connections page lists what exists and links here. It does not connect anything, because there is no project standing outside a project to connect it to.

Models available

Whatever the endpoint serves — you name the model when you configure the agent, and RAiN passes it through unchanged.

Choose per agent. A sequence can mix them — a cheap model for classification and a stronger one for the step that writes to a customer.

What it costs

Billed by OpenAI-compatible endpoint to the account the key belongs to. RAiN adds nothing and sees nothing beyond the tokens each run used, which the Execution tab shows.

When it goes wrong

It needs a base URL as well as a key, and the URL must be https. If the endpoint does not implement the chat-completions shape faithfully, runs fail in ways that look like model errors — test with one simple agent before building on it.