Download

Usage you can see, budgets that hold

See what every agent costs. Set a cap. Work stops on its own at the limit.

v2.2

release · May 27, 2026

  • Per-model cost on every run
  • Budgets per agent, per Space, per org
  • Work pauses at the limit and resumes when budget returns
  • A replayable audit log of every decision

You pay for the work an agent does, so you should be able to see it. This release makes usage and limits first-class.

  • Per-model cost. Every run is attributed down to the model that did the work, across Opus 4.7, Sonnet 4.6, Haiku 4.5, and Kimi K2.6.
  • Budgets. Per agent, per Space, per org. At 80% the Space is notified. At 100%, work pauses where it stands and resumes when budget returns.
  • Audit log. Every read, write, decision, and approval, in one replayable record.

Idle time, observed events, and human messages don't count against the meter. You pay for work, not for waiting.

All changelog entries