Manage model access and allowances

Model access decides what a caller may use. Allowances limit how much they may use.

Model access

Model access determines which public models and capabilities a caller may use. Gateway resolves it from the policies that apply to the organization, groups, user, App, and credential. A credential can narrow access, but it cannot expand access granted by the organization.

The Developer Portal shows the models available to your credential. Organization administrators manage organization model access and can review an App’s effective access. A model in the broader catalog is not automatically available to every organization or credential.

Use a public model ID from the Developer Portal or GET /v1/models. Provider-native IDs and internal route IDs are not user-facing values.

Allowances

An allowance controls how much a user or App may consume. Gateway can apply request, token, and estimated-cost allowances over minute, day, week, or month windows. The Developer Portal shows the allowance for your personal credential and its reset time. Organization administrators manage organization and App governance.

Each caller has one effective allowance. A direct allowance takes precedence over a primary group allowance, which takes precedence over the organization default. If several group allowances apply and no primary group is selected, Gateway uses the most restrictive one. Credential settings can narrow capacity but cannot expand organization governance.

Allowances can cover requests, tokens, and estimated cost over minute, day, week, or month windows. The Developer Portal shows the effective allowance, current usage, and reset time for a personal credential. Organization Administration shows the effective result for a person, group, App, or credential.

Usage history remains available after an allowance resets. A reset starts a new counter; it does not remove historical usage.

For grouped usage from coding tools, see Group usage by activity. Activity groups are a convenience view; raw usage events remain authoritative.

Selecting a model

Use an explicit public model ID when your application requires a specific model:

{
  "model": "your-public-model-id"
}

Use model: "auto" when both the request and organization support Gateway-managed selection:

{
  "model": "auto"
}

auto selects one eligible public model for the request. It does not bypass model access, capability restrictions, or allowances.

When a request is denied or limited

• 403 indicates that the credential is not authorized for the requested model, capability, or operation.
• 400 indicates that the request shape or requested capability is not supported by the selected transport, model, or effective limits.
• 429 indicates that an applicable rate, token, budget, or quota limit has been exceeded.
• 503 can indicate that the Gateway, control plane, or eligible provider route is temporarily unavailable.

Check the model list, current allowance, and reset time in the Developer Portal. Ask an organization administrator to review the App, policy, or allowance when a service needs different access.

See Inspect Gateway usage for request history, activity groups, breakdowns, and pricing health.