See every model each configured provider offers, its context window and price, and which models have no reachable credential behind them.

AI Inventory → Models is the catalogue of every model this appliance knows about, what it costs, what it can do, and whether a credential exists that can actually reach it.

What do I need?

Licence: Any
Role: admin_user

Prerequisites

At least one provider credential exists, otherwise the page shows the readiness banner and no model has a route.

Note

This page is read only. Models cannot be added, edited, enabled or disabled from it in 0.4, and there is no routing configuration here — model routing is a pane under Settings. What the page answers is what can this appliance call, and at what price.

Screenshot pendingAI Inventory Models with the model distribution charts above the model table/assets/screenshots/models@0.4.png

Read the inventory

Open the page

Select AI Inventory → Models.

Three charts sit above the table — Kind of Models, Distribution of Model Families and Model Usage — followed by the model table. The table's subtitle counts the models, providers, how many are active, and how many have no reachable credential.

Narrow to a provider

Pick a provider from the Provider filter, or type into Search for model....

Both filters are applied by the gateway, so the counts and paging follow them.

Read a row

Each row carries Name, Provider, Kind, Context, Input Cost, Output Cost, Features and Status.

Kind is Open Source or Proprietary. Features shows Vision for models that accept images and Streaming for models the gateway can stream. Costs are shown per million tokens, or Free.

Refresh after a credential change

Select Refresh.

The charts, the table and the routability check are re-fetched. The page does not poll on its own.

Deal with a model that has no route

Find the warning

Look for a No route tag on a model row.

Its tooltip reads This provider has no reachable credential — the model is listed but calls will fail. Add or test a credential under Credentials.

Fix the credential

Open Credentials, add a credential for that provider or select Open menu → Test connection on the existing one.

A credential that tests Verified ✓ clears the warning on the next Refresh.

Which models a user sees

The catalogue on this page is appliance-wide. What an individual sees when they pick a model is narrower, and is derived rather than configured:

The gateway collects the providers reachable through that account's own credentials.
It adds the providers reachable through every tenant group the account belongs to.
It lists the active models of those providers, sorted by display name.
Self-hosted red-team models are withheld from a normal_user — only administrative roles and the read-only viewer see them.

An account with no credential, direct or inherited, is offered no models at all. That is a credential problem, not a model problem: give the account a credential or put it in a group that has one.

Fields reference

Column	Shows
Name	The model's display name, with its API identifier
Provider	The provider that serves it
Kind	Open Source or Proprietary
Context	The context window, abbreviated — `128K`, `1.0M`
Input Cost	Price per million input tokens, or Free
Output Cost	Price per million output tokens, or Free
Features	Vision when the model accepts images, Streaming when responses can stream
Status	Active or Inactive. Only active models are offered to users
No route	The provider has no reachable credential; calls to this model will fail

Control	What it does
Search for model...	Substring match on the model name, applied by the gateway
Provider / All Providers	Filters to one provider
Refresh	Re-fetches the catalogue, the statistics, the usage chart and the routability check

When no provider is configured or none is reachable, a banner sits above the charts:

Banner	Meaning
No AI provider is configured yet	No credential exists. Add provider opens Credentials
No AI provider is currently reachable	Credentials exist but none answered its last test
Your provider keys are saved but unreachable — check egress/firewall, or use the Test action on the Credentials page.	The connected-deployment wording of the same state
This box is marked air-gapped, so cloud API keys won't work. Add a local model endpoint (Ollama or any OpenAI-compatible server) to enable chat and the AI help assistant.	The appliance is declared air-gapped; Add local model is the action

Verify

The table subtitle reports the provider and model counts you expect, and all active or a count of active models.
No row carries No route for a provider you have just configured.
Sign in as a user in a group that owns that provider's credential and open the chat. The provider's active models appear in the model picker.
Nothing on this page writes to the audit trail — it performs no changes.

If it fails

Error loading data. Please try again. — the model list could not be fetched. Check that the gateway is healthy and reload.
No results. with filters set — the search or provider filter matched nothing. Clear them before concluding the catalogue is empty.
No usage data available yet in Model Usage — no completed requests have been recorded yet. The chart fills once traffic flows.
A model is listed but a chat request fails — check for the No route tag, then test the provider's credential on Credentials.
A user reports an empty model picker — the account has no credential of its own and belongs to no group that has one. See Tenant groups.

Next steps

Credentials — the keys that decide which of these models can actually be called.
Token cost and savings — what the models on this page are costing.
MCP servers — the other half of the AI inventory.

Review the model inventory