|

Usage and Billing

Tenant-attributed metering and invoice exports.

Attribution dimensions

Every model call, embedding, search, and tool execution records project, environment, tenant, user, agent, installation, session, run, model, and provider. Pre-attribution rows (before your cutover) carry nulls rather than guesses.

Reading usage
  • GET /v1/usage — summary grouped by tenant, agent, installation, model, and day (90-day window max).
  • GET /v1/usage/events — paginated attributed rows for debugging and support.
  • GET /v1/usage/export?format=csv — invoice-quality export grouped by tenant and agent (5,000-row cap, truncated flag when hit). Narrow with tenant_id / agent_id.

Budgets and limits
  • Rate/concurrency (EMBEDDED_LIMITS, per tier): run creation returns 429 with Retry-After when per-scope rates or concurrent-run caps trip. Bursty tenants are isolated by fair scheduling — one tenant cannot starve the project queue.
  • Spend budgets: installations (budget: { max_spend_usd, window_days }) and tenants (metadata.budget) gate run creation and delegation with 402 budget_exceeded. Session turns are metered, not gated, to protect p95 latency.
  • Entitlements: embeddedAgents, embeddedSkillLibrary, embeddedSkillImport, embeddedRemoteMcp, embeddedBrowserPolicy, embeddedSubagents are tier flags; count caps (maxTenants, maxSkills, …) are the enforcement mechanism.

Billing your own users

Use rate plans (/v1/rate-plans) and end-user administration (/v1/end-users) plus the usage export to build your own billing UI. Cencori is not the merchant of record for your marketplace. Tiered/volume plans graduate on real month-to-date usage, including in-batch accumulation.

Did you like the content?