Introduction

Most models arrive on their own. Connect a model provider in Discovery and its catalogue lands in the Model Registry without anyone adding anything.

+ Add Model is for the gaps: a model served from a platform you have not connected, or one you want represented before its provider connection exists.

Where to find it

AI Governance > Model Registry > + Add Model

Access to AI Governance is granted through platform roles. Ask your workspace administrator, or see Settings > Governance > Roles for how permissions are assigned.

Before you begin

1. Check whether Discovery already covers it — If the provider is connectable, connecting it is almost always the better move — one connection brings in the whole catalogue and keeps it current. Adding by hand covers one model, once.

2. Confirm the platform is available for model registration — The + Add Model picker is narrower than the Discovery picker. See the Supported Providers Matrix.

3. Know the model identifier — Use the platform's own identifier rather than a friendly name. It is what appears in spans, and it is what makes the registry reconcilable against a provider invoice.

Add the model

Step 1: Open the Model Registry

Go to AI Governance > Model Registry. The count beside the page title is how many models are registered.

Step 2: Click + Add Model

A panel opens on the right, titled Add a model, with the subtitle Select the platform this model is served from.

Step 3: Select the platform

The list covers the platforms a model can be served from: AWS Bedrock, Anthropic, OpenAI, Snowflake, UnifyApps and Core42 Compass. It is shorter than Discovery's, because not every platform that hosts agents also serves models. Search or scroll, then select.

Step 4: Choose the connection

The panel becomes Select a connection for <Platform> and lists the connections that exist for it. Pick the one whose catalogue you want to import from.

Step 5: Pick the models to import

Under Model, the panel invites you to "pick one or more models from this connection to import". It is a multi-select list with a search box, Select all and Clear all.

This is an import picker, not a details form. You choose from what the connection actually serves rather than typing a model in by hand, which is why the identifiers always match the platform's own. If the connection returns nothing you will see No models found.

Step 6: Import

The button counts your selection, for example Import 3 models. UnifyApps confirms with Models imported successfully.

Result

The model appears in the Model Registry with the platform you selected, with you as its owner, and STATUS set accordingly. Until it is called, LAST SPAN shows a dash and ACTIVITY shows 0 spans — expected for a model that is registered but not yet in use.

Registration puts the model in your inventory. It does not govern it: apply a policy for that. See Apply a Policy to Assets.

Important considerations

1. Manual entries do not stay current — A hand-added model is a snapshot. When the provider adds or retires models, only a connection keeps up.

2. Watch for duplicates — If you add a model by hand and later connect its provider, you may end up with two rows for one model. Check before connecting.

3. Use the provider's identifier — A friendly name breaks reconciliation against spans and against the provider's own billing.

4. Registered is not reachable — Adding a model to the registry does not grant anyone access to it. This is an inventory, not a provisioning tool.