Compass FinOps

Compass FinOps shows AI spend, budgets, rate limits and month-end forecasts per role and per user, enforced through the Frontier llm-router with llm-d first.

Compass FinOps is a top-nav tab next to Admin that shows what AI costs, who is spending it, and who is on track to exceed budget. Access is gated by a viewer role.

AI Spend

Tile	Shows
Today	Spend and calls today
7 days	Spend and calls over the last 7 days
30 days	Spend and calls over the last 30 days
Month to date	Spend and calls this month
Escalations	Requests escalated to a paid tier

Export CSV downloads the data. Manage budgets links to Admin.

Budgets and rate limits

Budgets and rate limits are set per role and per user.

Field	Meaning
Monthly ($)	Monthly budget
Extension	Extra budget added for the current month
Spent MTD	Spend this month
%	Share of budget used
RPM	Requests per minute
TPM	Tokens per minute

A user budget overrides the role budget, and the most restrictive limit wins. Users and roles are picked from the identity provider (Keycloak today). Budgets can be extended.

Spend breakdown

Spend breaks down by model, app, role and user, plus a recent-calls table. Every table has sortable headers and a filter, and tooltips explain each field.

Enforcement

Budgets and limits are enforced by the Frontier llm-router. It sends requests to local models first, such as Gemma at $0 served through llm-d (a CNCF Sandbox project), and escalates to paid commercial frontier tiers when needed.

Forecast

Once 3 days of history exist, FinOps projects month-end spend. It flags principals projected to exceed their budget, along with the day they are projected to breach it.