{provider}/{model}.
Register a provider
Open the entity menu, choose Providers, and enter:- Name: Lowercase identifier used in every model reference, such as
openaioropenrouter. - Base URL: The exact OpenAI-compatible API root, including any required path prefix such as
/v1, without query parameters or embedded credentials. - API key: Encrypted before storage and never returned.
API: Register a provider (PUT /providers/{provider})
API: Register a provider (PUT /providers/{provider})
The provider name is the last URL segment. This example registers Poll
openai:GET /providers/openai until its deployment is applied. See Create or rotate a provider.Studio uses
POST /responses for the openai provider so relabeling and
evaluations preserve OpenAI reasoning and tool-call behavior. Other providers
must implement the OpenAI-compatible POST /chat/completions endpoint.Enable model IDs
Expand a provider’s model list to:- enable or disable known models in Studio selectors;
- add a custom model ID exposed by that provider;
- set optional input and output prices in USD per million tokens.
Built-in wandb-inference
Every entity has a system provider named wandb-inference. It uses the caller’s W&B key rather than storing another credential. Use it for W&B-hosted catalog models and fine-tuned artifacts.
You can explicitly register the same name to replace passthrough behavior with a stored credential. Deleting that override restores the system provider on its next use.