Workspace
Usage and limits
Track the compute your runs use, and the tokens they spend on your model keys.
Admins find usage under Settings → Usage. Usage is counted per billing period: a calendar month, in UTC.
Compute
Compute is the wall-clock time runs spend in a cloud sandbox, from when the sandbox starts working on a run until it’s done. Runs that never started aren’t counted. Answers from fast mode have no sandbox: their time counts at 1.5×, and the runs table marks them Fast mode · 1.5×.
The Overview shows:
- Current plan: your plan, its price, and when its included hours reset;
- Included in your plan: a bar for the compute used this period, with how much is left and any overage, and one for Codebot’s beta model credit while you have it;
- the compute used in the date range you picked, with the number of runs and the average run;
- a chart of compute over time, grouped by repository or by workflow, with a table view;
- Runs: every run that used compute, newest first, and whether it came from included hours or overage.
Runners
If your organization uses self-hosted runners, the Runners tab shows the time your runs spent on them. Runner time isn’t billed and doesn’t count toward your plan's compute. The tab shows:
- Runner time: claim to finish, like compute, with the number of runs and the average run;
- Queue wait: how long runs waited for a free runner, as the median and the slowest tenth (p90);
- Didn't start: runs that failed or were cancelled before any runner took them, for example because no runner was online;
- a chart of runner time grouped by repository, workflow or runner;
- Runners: runs, time, average wait and the most runs each runner had at once, against its concurrency;
- Runs: every run on your runners, with the runner and how long it waited.
Models
The Models tab shows this period’s model usage across your keys: tokens by kind (input, output, cache read), what that costs at your providers’ list prices, the cache hit rate, tool calls, and a table per model.
Model costs are billed by your providers, not by Codebot. See Model providers.
When you reach your limit
On the Free plan, runs stop when the included compute for the period is used up. New runs are recorded as cancelled with the reason, and on GitHub the pull request gets a skipped check that links to the usage page. Runs already in progress are allowed to finish. Everything starts again at the next period.
Keeping compute down
- Skip pull requests that don’t need a review with skip rules: drafts, bots, ignored files.
- Use Commits pushed only if you want a review on every push; Opened and Ready for review review less often.
- Set a monthly compute limit on busy Discord channels.