Usage
Track calls and costs per project.
The Usage tab shows what your project is consuming — a daily breakdown plus KPI sparklines, built from rated rollups merged with live events so the current day is up to date.
What's metered
Each billable engine call emits a usage event with a kind:
| Kind | Counts |
|---|---|
compile | Compilation runs |
retrieval | Hybrid / vector search |
ingest | Claim ingestion (contribute) |
ask | Grounded LLM answers |
function_invoke | Root Function executions |
branch_op | Branch operations (merge, diff, …) |
mcp_tool_call | MCP tool invocations |
query | Other reads |
Events accumulate in usage_events; nightly rating aggregates them into
usage_rollups (per project, per day, per kind) with a cost in cents. Only
successful (2xx) calls are metered — a failed proxy never bills.
If you serve many end-users under one project, each event is also tagged with
the end-user (X-TR-User), so you can break usage down per end-user — see
Multi-tenant SaaS.
Never charged for failures
Usage is rated nightly and failed runs are not charged. The figures you see are real signals from actual calls — nothing is estimated or padded.
For plans, quotas, and invoices, see Billing.