The model catalog

Build › Models is the model catalog: the open-weight LLMs, embedding models and rerankers your tenant can deploy, and the external models a cloud provider serves. Builders use it to see what is running; tenant admins deploy and undeploy from it.

Who can open it

Role	What they can do
Admin	Read the catalog, deploy a model for the tenant, undeploy it
Builder	Read the catalog and every model’s status
Auditor	Read the catalog and every model’s status
Reviewer	No access
End user	No access
Platform admin	Holds every capability: also adds models to the catalog, removes them, sets the LLMaaS offering, and sees the External tab

The catalog is shared by every tenant on the platform. A platform admin decides what is in it; a tenant admin decides what runs for the tenant.

What you see

The page opens with its header, Models, and one line: open-weight models to deploy for a tenant, and external models other providers serve.

A platform admin sees two tabs, Catalog and External. Everyone else sees the catalog alone. The External tab is described in External models.

In order, the catalog shows:

The tenant picker. It names the tenant whose deployments the statuses describe. It is hidden when you belong to one tenant.
The toolbar. A search box (Search models) that matches the display name, the model id and the family; a Cards / Table switch; and, for a platform admin, Add model and a More actions menu with Add by id.
The type filter. One tag per type, each with its count: LLM, Embedding, Reranker. Select any number. With a search or a filter on, a line reads Showing: n of total.
The models, as cards or as a table.

Cards are the default. The layout you choose is remembered in your browser, and ?view=table in the address opens the table directly.

Cards

Each card shows the model’s family, its display name (linked to the model’s page), its catalog id when it differs from the display name, and tags for type, Parameters, Offered as LLMaaS and, for an external model, Provider. An external model also shows its Data residency line. The card ends with its status and one button: Deploy when you may deploy it, otherwise Open.

Table

Column	What it holds
Model	Display name, linked to the model’s page, with the catalog id beneath
Type	LLM, Embedding or Reranker
Parameters	The parameter count, such as `8B` or `568M`. Sorts by magnitude
Status	The model’s deployment status for the selected tenant, with the reason beneath when it failed or is not served
LLMaaS	Offered when the platform offers the model as a managed endpoint
Actions	Deploy or Undeploy, and, for a platform admin, the catalog’s own actions

Statuses

Status	Meaning
Not deployed	Nothing runs for this tenant. Deploy is offered
Pending	The deployment was accepted and is being placed
Deploying	The model server is starting and its weights are loading
Active	The model is ready and the gateway publishes it to the tenant
Not served	The model is running, but the gateway is not publishing it. The reason is on the row
Failed	The deployment did not complete. The reason is on the row, and Deploy is offered again
Removing	An undeploy is in progress
Unknown	What is deployed could not be read
Served by provider	An external model. It is never deployed: the provider serves it

While any model is Pending, Deploying or Removing, the page refreshes every 3 seconds. No action is offered on a row while its status is changing or unknown.

Empty and error states

No models in the catalog. A platform admin is offered Add model. Everyone else reads that a platform admin adds the models tenants can deploy.
No models match. Your search and type filters exclude everything. Clear search and filters resets both.
The model catalog could not be loaded. Retry from the message.
What is deployed could not be loaded. The catalog is shown, but every status reads Unknown, not Not deployed. Try again reads the deployments again.

What you can do

Find a model

Go to Build › Models.
Type in Search models, or select one or more type tags.
Select a model’s name to open its page.

Deploy a model

Tenant admins deploy from the card, the table row or the model’s page. Deploy opens the deploy wizard. See Deploy a model.

Undeploy a model

Find the model. Its status is Active, Not served or Failed.
In the table, select Undeploy on its row. From cards, open the model’s page and use Undeploy there.
Confirm in Undeploy model.

The model stops serving for the tenant: requests to it fail and the tenant’s grant on it is revoked. Deploying it again starts from the beginning.

Curate the catalog (platform admins)

Add model opens Add a model from Hugging Face: search the hub or pick from the trending list, review the model card, then add it to the catalog. Deploy after adding starts a deployment for the selected tenant in the same wizard.
More actions › Add by id opens Add a model by its id: a model id, display name, type, family, parameter count and description, for a model the hub search does not reach.
Remove from catalog, on a row’s menu or a card’s menu, removes the model for every tenant. It is refused while any tenant still runs the model, and the model can be added again later.
LLMaaS offering… offers the model as a managed endpoint tenants subscribe to, and sets its GPU footprint, which decides how fast its use draws down a token budget.

Bringing packaged weights onto a cluster is covered in Model registry.

Limits

Rule	Value
Model types	LLM, Embedding, Reranker
Model id when added by id	One name, or `organization/name`. No spaces, and none of `?`, `#`, `%` or `\`
Status refresh while a deployment changes	Every 3 seconds
Remove from catalog	Refused while any tenant runs the model