Model visibility

Choose which models a model list shows. Turn listing off, show only some providers or models, hide models by name, or hide undated aliases, for everyone or for an app, a team or a person. Model visibility is a Policy Engine card. With the engine off, every list shows the models a caller may use, as before.

Visibility is not access

Hiding a model only takes it off lists. It never blocks a call:

• A hidden model still works by name: in requests, in GET /v1/models/{id} and through Add a model by name.
• To stop a model from being used, reject it on the Allowed Models card. GET /v1/models applies that card first, and never lists a model it rejects, not even to callers who ask for hidden models.

You want to	Use
Keep a model out of pickers and /v1/models, while code that names it keeps working	Model Visibility
Stop a model from being called	Allowed Models
Stop someone from sending requests at all	Gateway Access

Lists it applies to

List	Who reads it
GET /openai_compatible/v1/models and GET /anthropic_messages/v1/models	Coding agents such as Cursor and Claude Code, chat front ends such as Open WebUI, and the OpenAI and Anthropic SDKs, calling with an app's Quilr key
Get available models in an app's providers, and Check key and list models, Get available models or List models when you add a provider or models	Admins in the console

Other lists are unchanged: Policy Engine pickers, the Models pages and the playground show every configured model.

The two kinds of list start from different places. GET /v1/models lists the models the caller may use, after Allowed Models. Console lookups list what the provider key can reach, so you can choose models for a provider; Allowed Models does not apply there, only these rules.

Scopes match like this:

• People and Smart group: on GET /v1/models, the identity the request carries (a JWT, or X-User-Email when the app accepts it; see Identity and network trust). In the console, the person signed in.
• Application and App tag: on GET /v1/models, the app whose key made the call. In the console, only Get available models on a provider an app already has carries the app. Creating an app, entering new credentials or adding a provider in Models carries none, so those rules do not apply there.

Model Visibility card

Go toQuilrAI consoleGovernPolicy EngineLLM GatewayModel Visibility

The card sits next to Allowed Models. Start from Quick add or from a control's add button:

Quick add	What it sets
Hide preview models	Hide with the model pattern *preview*
Only show approved providers	Show only with the providers you pick
Turn off model listing	Model listing set to Off
Let admins list hidden models	Hidden models set to On request for the people or Smart groups you pick

Each opens New model visibility rule. Choose who it applies to (everyone, or People, Smart group, Department, IdP group, Application or App tag), then what their model lists show:

Setting	What it does	Not set
Model lists	Both, API (GET /v1/models) or Console	Both
Model listing	Off: lists show nothing. GET /v1/models returns an empty list, and the console asks for model names.	On
Show only	Tick Show only some models and add grants. Only models a grant matches are listed.	Every model
Hide	Tick Hide some models and add grants. Models a grant matches are left out, even where a show-only grant matches them.	Nothing hidden
Undated aliases	Hide undated aliases: leave out gpt-4o when the same provider also lists gpt-4o-2024-08-06.	Listed
Hidden models	Never, On request or Always: who still sees hidden models, marked as hidden. See Show hidden models.	Never

Add to draft puts the rule in the shared draft, and it applies once you publish a revision. The card lists each rule's grants, for example azureopenai · gpt-4o*. Editing a rule reopens the same form, or the full editor for a rule the form cannot show.

Grants

A grant describes a set of models. Fill in one or more of its fields, and use Add grant for another set.

Field	Matches	Example
Providers	The provider type	openai, azureopenai, bedrock
Provider labels	The label saved on a provider	prod-east
Model patterns	The model ID. * matches any run of characters and ? one character.	gpt-4o*, *preview*

• A grant matches a model when every field you filled in matches. Inside one field, any value can match.
• Providers take the exact provider type: openai is OpenAI Chat completions, so an OpenAI Responses provider needs openai_responses. The field suggests the types your apps use.
• Matching ignores case. A rule holds up to 16 grants, and a field up to 128 values of up to 256 characters.
• A pattern matches the whole ID: gpt-4o matches only gpt-4o, while gpt-4o* also matches gpt-4o-mini and gpt-4o-2024-08-06. Use *preview* for "contains preview".
• A provider label matches only a saved provider. A provider you are still adding has no label yet, so label grants never match it.

How rules combine

The gateway builds each list in this order:

1. Allowed Models removes the models the caller may not use (on GET /v1/models only).
2. Model listing set to Off by any matching rule empties the list.
3. Show only: when matching rules have show-only grants, only the models that match one of them stay.
4. Hide: models that match any hide grant are left out, even where a show-only grant matches them.
5. Undated aliases: hidden when any matching rule hides them.
6. Hidden models decides whether hidden models come back, marked.

• Grants add up. Show-only and hide grants from every matching rule combine, whatever their priority. A higher-priority rule cannot take back another rule's show-only grant, so hide what should not be shown instead.
• Off and hidden always win. Model listing Off and Hide undated aliases apply when any matching rule sets them. To exempt a narrower scope, such as admins, set Hidden models to Always for it.
• Hidden models follows priority. The highest-priority setting wins, and at one priority the stricter one wins (Never, then On request, then Always). The form gives a rule for a person, Smart group or app a higher priority than a rule for everyone.
• One model, two providers. When two providers offer the same model ID and only one of them is hidden, the list still shows the model, under the provider that is visible.

A model visibility rule holds only model visibility settings, plus its name, priority and risk level. In Advanced policies, validation rejects a rule that also sets other controls, because other requests would then skip the whole rule. It also rejects conditions a model list never carries: the requested model, prompt text, tools or detected data. It warns when a rule matches on the provider (use a grant's Providers instead), on the source IP of a console lookup (that is the console server's address), or hides * (set Model listing to Off instead).

Show hidden models

Hidden models decides who still sees what these rules hide. Hidden models come back marked. A model that Allowed Models rejects never comes back on GET /v1/models.

Hidden models	GET /v1/models	Console lookups
Never (the default)	Never listed	Listed only when someone who can view LLM Gateway policies ticks Show models hidden by policy
On request	Listed when the caller asks with ?include_hidden=true or the X-Quilr-Include-Hidden: true header	Same as Never
Always	Always listed	Always listed

Without On request or Always, include_hidden is ignored: the caller gets the normal list and no error. Each listed hidden model carries "quilr_visibility": "hidden":

curl "https://guardrails.quilr.ai/openai_compatible/v1/models?include_hidden=true" \
  -H "Authorization: Bearer $QUILR_KEY" \
  -H "X-User-Email: admin@acme.com"

{
  "object": "list",
  "data": [
    { "id": "gpt-4.1", "object": "model", "created": 0, "owned_by": "openai" },
    { "id": "gpt-4.1-preview", "object": "model", "created": 0, "owned_by": "openai", "quilr_visibility": "hidden" }
  ]
}

In the console, Show models hidden by policy appears once a lookup hid something, and only for people whose role lets them view LLM Gateway policies. Hidden models then show as gpt-4.1-preview · hidden by policy in an app's providers, or with a Hidden by policy tag when you add a provider or models.

What people see

• API callers get fewer entries, in the usual format. With listing off, data is an empty list. A hidden model still answers GET /v1/models/{id}, while a model the caller may not use returns 404 (model_not_found in the OpenAI format).
• Console lookups add a line such as "3 models hidden by policy." Hidden models are dropped before the console checks each model with the provider, so they are never called.
• Nothing left to list. With listing off, or when no show-only grant can match the provider, the console does not contact the provider. It shows "Model listing is turned off by your organization's policy. Add models by name." or "All models from this provider are hidden by your organization's policy. Add models by name." Type models into Add a model by name (or Add by ID): each is still checked against the provider, which also checks the key.
• If the policy cannot be read, lists show every model the caller may use, as they would without these rules. Visibility is presentation, so it fails open.

Scenarios

• Tidy lists for everyone. Hide *preview* and undated aliases for everyone:

tidy_model_listsrequest

runs on request·priority 500

Whenalways

Then

Hide models→*preview*

Hide undated aliases→true

• Approved providers only. Show only azureopenai for everyone. A team that also uses Anthropic gets its own show-only grant for anthropic_messages on its Smart group, which adds to the first.
• No model list for one app. Set Model listing to Off for an Application whose users pick models from your own approved list. Requests that name a model keep working.
• Admins see everything. Set Hidden models to On request for a Platform admins Smart group. They call GET /v1/models?include_hidden=true and see every model they may use, with hidden ones marked:

platform_admins_list_hidden_modelsrequest

runs on request·priority 900

WhenSmart groupsincludes (ignoring case)Platform admins

Then

Show hidden models→on_request

• Coding agents see one family. For an App tag shared by your coding-agent apps, with Model lists set to API (GET /v1/models), show only azureopenai · gpt-4o*. Console lookups for those apps are unchanged.
• Retire a provider's copies. Hide the provider label legacy-openai. Models that another provider also offers stay listed, under that provider.

Related

• Gateway access and allowed models
• Providers and models
• Identity and network trust
• Policy Engine overview