Setting quotas

What a quota measures

Administration → People & Usage → Quotas. Each quota caps input tokens, output tokens, spend (USD) or requests over a calendar day, week or month, or a rolling 24 hours, 7 days or 30 days. It applies to a person or a team, and covers all models or one model, managed model or router; a managed model or router counts everything sent to it, whichever model served.

Warning alerts go out at 80% and 95% of the limit unless you set your own percentages (up to 10); a breach alert always fires at 100%.

A window that needs more usage history than Janus keeps (JANUS_USAGE_RETENTION_DAYS, 365 days by default) is refused, because it would undercount. Team quotas need that history even for calendar windows.

Service tokens can be capped too, from the Quotas page or with subject_type service_token in POST /api/v1/admin/quotas.

Rate limits (Quotas → Rate limits → New rate limit) cap requests per minute on an endpoint. Applies to: people (everyone or one person) or a service token; a service-token rule throttles only that integration. Rate limits are enforced while the per-endpoint rate limits feature flag is on.

A Janus Chat reply counts as one request toward rate limits and request-count quotas, however many calls it makes (tool rounds, deep research, media, title). Tokens and cost count for every call. With Compare, each model’s answer counts as one request.

Set limits with headroom

The default breach behavior lets an in-flight request finish, so a single large streaming completion can overshoot the limit. Choose a limit that tolerates one maximum-size request beyond it.

“Cut the stream immediately” stops a stream as soon as the limit is reached, counting the stream’s own tokens, so the overshoot is at most the last chunk — but it cuts users off mid-answer. Reach for it only when a hard ceiling matters more than the experience.

Team quotas

A team quota measures every member’s usage together. Combined with a per-user quota it produces a shared budget with an individual cap inside it.

Delegating to team leads

Each team’s page in Administration → People & Usage → People → Teams has a “Team leads may edit quotas” checkbox under Administrator controls. When an administrator ticks it, that team’s lead gets a quota-management panel on their team page and can add, edit (limit and whether to cut streams) and remove quotas for that team only — never for other teams, individuals, or global settings. Leads cannot create quotas while license creation is paused, and cannot create or change quotas of a deleted team.

Every change a lead makes is audit-logged with their identity, exactly like an admin edit. Untick the checkbox to withdraw the delegation at any time; existing quotas keep enforcing either way.