Admin Settings → Models controls the workspace model library — the list every user sees in their model selector. Curating this list is one of the highest-impact admin tasks: fewer, well-chosen models make users faster and clusters calmer.
Manage Models
The Manage Models view shows all models discovered from your connected endpoints (Connections):
| Action | Effect |
|---|---|
| Select models that appear for every user (Default Models) | The curated set shown to everyone by default. |
| Show Model / Hide Model | Hide a model from the selector without deleting anything. Hidden models stay available to admins for testing. |
| Drag to reorder the workspace library | Order models by preference — the first ones appear at the top for users. Put your recommended default first. |
| Export | Save the model configuration for reuse or backup. |
| Reset All Models | Return the catalogue to its discovered defaults. This deletes custom entries — confirm before using. |
Note:
"What models should we expose?" — pick 3–6 that cover the use-cases (a fast general model, a strong reasoning model, a vision model, a coding model). A 40-model selector slows every decision.
Pulling and removing models
Where your deployment allows self-service model acquisition:
- Requesting a model by name from the Repository or model selector triggers a pull — the model is downloaded to the cluster and becomes servable.
- Removing a model frees its disk space; check Cluster Manager → Storage afterwards.
- Pulling large models consumes bandwidth, disk and memory — approve as an infrastructure change in production environments.
Model limits and access
| Control | Where |
|---|---|
| Which groups may use which models | Permissions |
| Cost reporting per model | Model Pricing |
| Quality per model | Evaluations leaderboard |
| Per-model defaults (parameters, persona) | Model editing — see Models Overview (user) |
Troubleshooting
| Message | Meaning |
|---|---|
| "No models found" | Connections aren't configured or endpoints are down — check Connections. |
| "No inference engine with management support found" | The connected model service doesn't offer management operations in your configuration. |
| Users can't see a model you added | It isn't in Default Models, is hidden, or their group lacks permission. |