Prices and limits are read from one config on our side and rendered on the pricing page,
the product page and your billing tab — so they cannot disagree with what you are charged.
What is not metered
Runs. Evals are bursty and CI-shaped, and metering them would show you a frightening number on the exact day you are evaluating us. Reporting is never rate-limited and no tier restricts it — see Reporting. Seats. Unlimited from the first paid tier. Per-seat pricing punishes the agency with six developers and one client project, which is precisely the shape of team this is for.Retention deletes less than you think
This is the part most worth reading carefully, because the natural assumption is wrong. Only per-sample detail ages out. Runs, their per-row scores, and baselines are kept forever, on every tier including free.
“Is this worse than it was three months ago” is the entire value of the product — deleting
the run would delete the answer. What ages out is the verbatim model input and output
hanging off it, which is both the bulk of the storage and the whole of the compliance
surface.
In practice: on the free tier your trend line still goes back to your first run a year
later. What you lose after 14 days is the ability to open a row from back then and read what
the model actually said.