Burn | track your AI coding spend

Built for the way you spend

Three numbers, not a report

Live burn rate

$/hr right now, projected to midnight against your typical day. You'll know you're running hot before the bill does.

Per-model breakdown

Opus, Sonnet, Haiku: see exactly which model is burning the budget, and how much of that is just cache read/write.

Cost per shipped PR

Weekly spend divided by merged PRs. It's the number that tells you whether the spend is paying for itself.

Burn settings screen listing connected Git hosts

Multi-provider by default

Claude and Codex, one number.

Switch the header chip to see one provider's spend or everything combined. Same view shows how close you are to a weekly or monthly plan cap, so you're not also juggling a separate dashboard or a spreadsheet.

PRs read from GitHub or any self-hosted Forgejo, Gitea, or GHES instance. Add hosts from Settings; tokens stay in Keychain.

Stop guessing what today cost.

Free. Runs in your menu bar. Reads local usage logs, checks your plan allowance where needed.

Download for Mac

or brew install maferland/tap/burn

FAQ

Where does the cost data come from?

Burn reads Claude Code's and Codex's local usage logs directly on your Mac and prices them against each provider's published rates. Checking usage against your plan's weekly or monthly allowance uses your provider account; everything else stays local.

Does it work with self-hosted Git?

Yes. Add any GitHub-compatible host (Forgejo, Gitea, GitHub Enterprise) from Settings alongside github.com. Tokens live in Keychain and never show up in the UI.

What's "cost per shipped PR"?

Your weekly (or monthly) spend divided by the number of PRs merged in that window. It's rough, but it's an honest read on whether the spend is turning into shipped work.

Is it free?

Yes.