Skip to main content
The Model catalog lists every model your organization’s provider credentials can serve, with its price and what it was used for this month. Owners and admins decide which models are approved, and can make the AI Gateway serve approved models only, on every key. Model catalog, under Govern, AI assets in the sidebar, is for owners and admins when your organization has the AI Gateway.

​
What each row shows

Each row is one model of one provider: gpt-4.1 through OpenAI and gpt-4.1 through Azure are two rows, each with its own status. A model your traffic used this month is listed even when your credentials no longer list it. The four figures above the table count the models by status. Filter by Provider, by Status (Deprecated too) and by Use (Used this month or Not used).

​
Decide about models

1

Choose the models

Tick one or more rows, up to 200 at once. The bar above the table says how many are selected.
2

Approve, review or block

Choose Approve, Review or Block, and add a note if you want one on the record, up to 500 characters.
Choose a row to open its detail: its prices and context window, the AI systems that used it this month with their requests and spend, and every decision about it, the newest first, with who made it and the note. Decide changes its status from there, and View requests opens the request log for that model. A status changes nothing on its own while Approved models only is off: it is a record of what your organization decided.

​
Approved models only

Turn on Approved models only at the top of the catalog and the AI Gateway serves approved models only, on every key: AI system keys, organization keys and people’s own keys, including a request that sends its own provider key. A model that is Under review, Blocked or Not reviewed is refused before it reaches the provider. It is off until an owner or admin turns it on. Before it turns on, Altrum asks, and names each model used this month that is not approved, with its requests, since those requests would be refused from then on. A refused request gets status 403: The request shows in the request log as refused, and the catalog counts it in that model’s Requests. Approve the model and the next request is served. While it is on, every decision reaches the AI Gateway at once. If the AI Gateway cannot be reached, the decision is not saved either, so the catalog never says Approved for a model that is refused:
The AI Gateway could not be updated, so nothing changed. Try again shortly.
Turning it off lets every key use the models its own list allows again.

​
How it combines with each key’s models

Approved models only works on top of what each key may use, set under Models and routing: a model must be on the key’s list and approved. A model name matches the approved one exactly, ignoring case, for the provider the request goes to.
  • GET /v1/models on a key lists only the models the key may use that are approved.
  • A request that names no model, such as an MCP tool call, is not affected.

​
Where else a status shows

Discover lists each model used in the range that nobody has decided about. Review opens the Model catalog. An AI system’s Bill of materials, on its Overview, shows Under review or Blocked beside a model it used in the last 30 days that has that status. Approved and not reviewed models show no badge.

​
Who can do what

Deciding and switching are limited to 120 changes an hour for each person.

​
Recorded in the audit trail

Each decision adds an entry for each model, Decided about a model in the Model catalog, and turning Approved models only on or off adds Turned approved models only on or off. See the audit trail.