Skip to main content
Use Factory → Analytics to compare tasks, models and stages. Use the ledger when you need usage for a run or a UTC day. Use one scope at a time: --run includes the run’s pinned per-run caps; --day aggregates workspace usage across factories and does not return caps:

Read the numbers

A subscription-backed run can show model usage even when no model API cost was billed. It can still incur sandbox compute cost. Look at By model for the model mix and By stage to see whether planning, building or reviewing used the most resources.

Set a budget

Merge this block into your factory.yaml and preview the push:
  • Active runs: excess work queues. Waiting for a gate is not the same as finishing a run.
  • Sandbox slots: a full workspace sandbox cap queues a launch or wake until a slot is free.
  • Sandbox hours: the cap includes committed time, including active reservations. A new reservation can be refused before all of that time is used.
  • Dollar caps: checked against recorded spend. A run can pause after a charge takes it past its cap; don’t treat the configured number as an exact billing cutoff.
Leave out budget to use the defaults ($50/run, $300/day). Once you supply budget, an omitted or null member removes that dollar cap. This doesn’t remove the time or slot limits. Runs use the configuration version they started with, so a higher limit in a new push doesn’t rewrite the cap on an old run. See the field reference.