What counts toward monthly monitoring usage?
One accepted model-call record counts as one log. Data usage is the byte size of the accepted JSON record after server normalization, using decimal MB and GB. Retries with the same log ID are counted only once. Deleting or expiring a stored log does not restore monthly usage.
When do monthly limits reset?
Each organization shares one monthly log, data and evaluation allowance. Usage resets on the first day of each calendar month at 00:00 UTC. Annual subscriptions use the same monthly limits; they do not provide the full year’s allowance at once.
What happens when a monitoring limit is reached?
Records that would exceed either the monthly log or data limit are not accepted. Existing prompt retrieval and your app’s direct model calls continue. Stored logs also have a per-prompt count cap and a metadata retention period.
How are evaluations counted?
One sample scored by one evaluator counts as one evaluation. Queued evaluations reserve that usage before running. Unscored work is released if a run fails or is canceled. Manual and automatic runs and calibration sample scoring share the monthly allowance; drafting or revising a rubric does not consume evaluation units. A manual run can include up to 1,000 logs. Automatic evaluation is available on Pro.
How do shared member limits work?
A personal workspace has one owner. Convert it to a shared workspace before inviting members. Free supports up to 5 members including the owner. Team and Pro have unlimited members. There are no per-member fees on any plan.
What does metadata retention include?
Monitoring metadata is retained for up to 7 days on Free, 30 days on Team and 90 days on Pro, subject to the per-prompt log count cap. Input and output payloads have a separate retention policy, with a 30-day default. A 90-day metadata window does not mean payloads remain available for 90 days.
Can I buy extra usage?
Additional usage packs are not available for purchase. There are no automatic overage charges.