Workspace LLM Collection

Understand how to authorize specific models for your workspaces and manage workspace-level toggles.

            Provider
               ↓
     Global LLM Collection
               ↓
    Workspace LLM Collection   ← You are here
        ├──────────────┐
        ▼              ▼
  Direct Model    Deployment
        \              /
         \            /
          ▼          ▼
         Virtual API Key
               ↓
        Gateway Request

By default, newly created workspaces have access to no models. Before models can be used within a workspace, they must be explicitly authorized from the Global LLM Collection.

This boundary provides clean isolation for separate environments (e.g., development, staging, production) or business units.

Whitelisting Models

To manage models in your workspace:

Navigate to the Model Whitelist (/workspaces/[wsId]/models) page in the workspace sidebar.
Click Manage Models in the top right.
Search or scroll through the global registry list, and click Whitelist to pull the model into your workspace.

Once added, the model's properties (RPM limits, capabilities, and provider configurations) are inherited from the global settings.

Active States

Models within a workspace have two distinct control levels:

1. Whitelist Membership

Adding or removing a model from the workspace. Removing a model completely revokes its access, and it can no longer be used as a target for deployments in that workspace.

2. Enabled/Disabled Toggle

Each whitelisted model has a status switch:

Enabled: The model is active and can process traffic or be used by routing deployments.
Disabled: The model is temporarily deactivated. Any deployment targeting this model directly will fail (or fallback to an alternative model if configured), and it will be excluded from active load-balancing pools.

Use cases for deactivation

Temporarily disabling a model is useful when:

Performing maintenance or rotating API keys for a specific provider.
Controlling budget spikes by turning off expensive models (e.g. gpt-4-32k or claude-3-opus) during non-critical phases.
Testing fallback behavior without deleting the model configuration.

Next Step: Deployments

Once a model is whitelisted, it can either be accessed directly using its registered alias name or attached to a virtual Deployment for routing, load balancing, and runtime processing.

Deployments can only reference models that have been whitelisted into the workspace. For detailed instructions on routing strategies and advanced configurations, see the Deployments documentation. To pre- or post-process payloads running through these models, see Runtime Modules.

Role-Based Access Control (RBAC)

Managing the Workspace LLM Collection requires specific permissions (see Roles & Permissions):

View Models: Allows members to view the list of whitelisted models.
Manage Models: Allows members to open the whitelist manager, add models from the Global Registry, or toggle their status.