Models

/models is the catalog of routable model groups and (for admins) the place to wire providers.

Concepts: Models and routing.

Tabs

Tab	Who	What you do
Model catalog	Everyone	Search and filter model groups, open the detail drawer
Access groups	Everyone	Model bundles granted to keys; writable platform admins create, edit and delete them
Provider credentials	Admin	Create, test or delete credentials (AES-256-GCM or akv: vault refs)
Add deployment	Admin	Wizard: provider + catalog model, model group, credential, weight
Pricing	Admin	Org prices vs catalog; set prices; reload the catalog
Health	Admin	Deployment health checks
Fallback chains	Admin	Ordered fallback chains across model groups

Tab badges count credentials that need attention (Provider credentials), model groups missing a price (Pricing) and failing deployments (Health). The selected tab is kept in the URL (for example /models?tab=pricing), so links open on it.

Stat cards: Total models, Active deployments (with a failing count), Providers connected, Fallback chains. Writable platform admins also get Add deployment in the page header, which opens that tab.

Model catalog

Search by model group, provider or provider model, and filter with the provider chips (All providers plus one per provider, with counts). Columns: Model, Provider, Context, In / out per 1M (org price), Deployments, Status (Active, Degraded, Retiring, Disabled).

Row menu: View details. Writable admins also get Test model, which runs a health check on every deployment of the group.

The detail drawer has two tabs:

• Details: context window, org price, catalog suggestion, deprecation, capabilities, fallback chain, and deployments with health and 7-day average latency. Writable admins can Edit, Disable / Enable or Delete each deployment.
• Usage: all-time requests, tokens and spend; spend per day for the last 30 days; top keys on this model. Non-admins see only their own spend and keys.

Writable admins get Test model, Set price and Set fallback in the drawer footer.

Access groups

Each card lists the group's models, marked Deployed, Deployments disabled or Not deployed, and how many keys use the group. Writable platform admins can add a group with New access group and use a card's menu for Edit group or Delete group. Edits apply to new approvals; existing keys keep their snapshot of the models. The same groups are managed under Access & Tiers.

Provider credentials

Create credential adds one. Each credential shows how many deployments use it and a status: Valid, Needs attention (a deployment using it failed its last health check), Not tested or Unused. Test runs a health check on every deployment that uses it.

Add deployment

Pick a provider and a catalog model, then set Public model identifier (what callers send as model), Provider model / deployment, Credential and Weight / priority. Test connection sends a minimal request to the provider. A summary shows where the group routes, the credential, context window and price.

There are no price fields: if the model group has no org price yet, the catalog's suggested price is applied. A deployment that repeats the same provider model, group and credential is refused. With no credentials yet, the tab links to Go to provider credentials.

Pricing

The Price catalog card shows when the catalog last synced and how many models it holds; writable admins can Reload catalog to import catalog prices and context windows. Org pricing lists every model group, including deployed groups without a price.

• Filter: All, Missing a price, Differs from catalog
• Columns: Model group, Provider, Ours / 1M, Catalog / 1M, Check (Matches catalog, Differs from catalog, No org price, Not in catalog), Updated (when, and by whom when known)
• Writable admins: Set pricing, and per row Apply catalog (when a catalog price exists and the org price is missing or differs), Set price or Edit

Health

Model health checks each deployment with a one-token request. Summary tiles count Healthy, Failing and Not checked yet deployments. The table shows each deployment's status, Avg latency · 7d and last check. Writable admins can Run all checks or check a single deployment.

Fallback chains

Each chain shows its routing order. Writable admins can add a New fallback chain, Edit order or delete a chain. A shortcut offers Add a chain for the busiest enabled model group that has none.

First-boot bootstrap

If OPENAI_API_KEY is set and no credentials exist, the first portal sign-in creates an openai-default credential and starter deployments. If that key was empty or stale, delete the credential here and add a new one.

Role restrictions

Non-admins only see the catalog and access groups. Platform admins see all seven tabs; view-only platform admins see them without write buttons. Mutations require a writable platform admin.