The suite · Actuals
What did your team actually spend, and was it worth the seats?
Subscription AI seats hide their economics: a flat fee, no meter, no way to see which seats earn their price. Actuals takes a usage export from OpenRouter, Claude Code, or Codex and reprices every seat's real consumption at verified metered rates, so each fee has a number next to it.
1 · Export
Drop in the usage CSV your vendor already gives you. No agent, no API keys, no integration work.
2 · Reprice
Each seat's tokens are repriced at the metered catalog rate for the same model: cache reads at cached rates, unknown models excluded and listed, never guessed.
3 · Verdict
Every seat lands in a band: keep (heavy use, seat is cheap), near breakeven, review, or move to metered (the seat costs more than the usage).
The sample is real generator output on synthetic usage for an 8-seat team: fees vs metered-equivalent, seat by seat.
Where this is today
The seat report runs now and we operate it with design partners by hand: you send an export, you get the report and a walkthrough. Self-serve upload, scheduled re-runs, and policy alerts (budget thresholds, model-mix drift, idle-seat flags) are the roadmap, built with the design partners, in that order.
The repricing rests on the same verified catalog and first-party benchmark runs as the Budgets side; seeEvidence for how those numbers are made.