Configure AI resources and the workspace assistant

Configure the provider and model resources used by your workspace, or select the model for its knowledge assistant. Keep these responsibilities distinct: a provider describes an available endpoint and credential reference, the model catalog describes models, and a versioned model configuration describes how a workload uses one.

Before you start

Confirm the selected namespace and workspace, the approved provider endpoint, intended model and data-handling requirements. Provider administration requires namespace administration access. Model catalog changes require provider.model_catalog.manage. Workload asset authoring and workspace assistant administration have their own permissions; one does not grant the others.

Store credentials through Connections and secrets. Select a credential reference in configuration; do not paste secret values into names, descriptions, screenshots or support messages.

Configure a provider

1. Open Connect → Models → Providers. Confirm the namespace before editing a provider shared by its workspaces.
2. Select Add provider, or Edit on an existing provider.
3. Complete Slug, Display name, Provider type and Base endpoint using the values approved for your deployment.
4. Choose Credential environment, then Credential secret from the eligible secrets in the selected workspace. A secret saved for all environments is not the same as an environment-specific credential binding. If the credential belongs to another workspace, switch there to change its binding.
5. Review whether the provider should be Enabled, then select Add provider or Save changes.
6. Read the credential verification outcome. If it is invalid or unverified, resolve the displayed reason; a saved provider record alone does not prove the credential works. Select Done when offered and inspect the provider row.

Use Discover models to inspect the models reported by a provider. Discovery and Reconcile are separate from configuring a workload or activating a release. Provider availability, credential validity and model eligibility all affect whether a workload can use the resource.

Changes to a shared provider can affect other workloads. Before disabling or removing one, identify its current uses and follow your organization’s change process. Resolve any active-use refusal instead of bypassing it.

Manage model catalog entries

1. Open Connect → Models → Model catalog and inspect the model’s provider, capabilities, context limits and available cost metadata.
2. If authorized to maintain the catalog, select Add model, choose Provider and Model, and enter the supported metadata. Keep token limits and pricing units consistent with the field labels.
3. Select Add model or, when editing, Save changes, and confirm the resulting row.

Catalog metadata helps selection; it is not a benchmark result or a guarantee that every listed capability works in your deployment. Disabling a model removes it from the served inventory and may affect selection by existing consumers. Check its uses before changing availability.

To use a model in a workload, open Build → Assets, create or edit the appropriate Model Config asset, and select the approved provider/model settings offered there. Save a version, bind the intended version to the release, and test and evaluate it. A catalog entry or saved asset is not an active release. Manage assets and versions explains the version lifecycle.

Configure the workspace knowledge assistant

You need workspace.manage in the target workspace. These settings affect the workspace’s assistant, rather than every workload using the namespace’s providers.

1. Open Administer → Workspaces, select the workspace, and find Knowledge assistant.
2. Choose Provider, then choose Model. Selecting the model saves the provider/model configuration immediately; there is no separate Save button.
3. Use Enable assistant or Disable assistant to set availability. This switch also saves immediately. Disabling it makes the assistant unavailable to workspace members.
4. Check the displayed state and any readiness explanation. After setup or re-ingestion completes, open Home and confirm availability. Follow Ask your first question when you are ready to submit a query.

A provider can be listed without a credential or while disabled. Choose an eligible provider with a bound credential; changing the model alone does not fix provider readiness. An in-progress knowledge refresh may need to finish before the assistant is ready.

If configuration does not work

For specialist integration details, use Technical provider reference and Technical model reference. For related release and model policies, see Review governance and evidence.