Burnbar
Privacy Policy
Burnbar exists because measuring AI token usage should not require handing over your prompts, your code, or your browser secrets. This page states exactly what we collect, what we never touch, how long anything is kept, and how to remove it. Every claim here is reviewed against the actual upload, storage, and retention code.
Last updated: May 2026
What we collect
Burnbar runs entirely on your Mac. The only data that would ever leave your machine is a small daily usage rollup you upload to the leaderboard after you opt in. The public leaderboard is rolling out soon. This is the upload model it will use once live — each uploaded row contains exactly these four fields and nothing else, and the server rejects any extra key:
- date
- The calendar day a usage total belongs to (YYYY-MM-DD).
- provider
- Which CLI the usage came from — only "claude" or "codex".
- tokens
- A single aggregate token count for that day and provider.
- cost_usd
- The estimated US-dollar cost for that day and provider.
Because the leaderboard is public, each row is tied to your GitHub account. This identity is not part of the uploaded row — the upload is authenticated with your GitHub token, and the server reads your public identity from it:
- github_id / github_login
- Your public GitHub numeric id and username. Not part of the uploaded row — the upload is authenticated with your GitHub token and the server reads these from it to label your row on the public leaderboard.
The leaderboard supports two providers only — Claude Code and OpenAI Codex. No other tool is read or uploaded.
What we never collect
The following are never read, never stored, and never uploaded. This is enforced in code: Burnbar parses only token counts, models, and timestamps from local CLI logs, and the leaderboard server validates every upload against a strict four-field allowlist before it is saved.
Prompt and response content
Never. We skip message.content / message.text, first_user_message, preview, title, and history.jsonl entirely. The text of what you ask or what the models reply is never read, stored, or uploaded.
Filesystem and repository identifiers
Never. We never upload cwd (working directory), git_branch, git_origin_url, git_sha, project directory names, or rollout paths.
Machine identifiers
Never. Cross-device totals are reconciled on your Mac before upload, so per-machine ids never leave your device and never reach the leaderboard.
Raw model names
Never. The upload carries a single per-provider token total, not which specific model (for example a dated Opus or Sonnet build) produced it.
Browser data
Never. Burnbar does not read browser cookies, Local Storage, IndexedDB, or any browser-managed secrets. This is the core reason Burnbar exists.
Third-party Keychain items
Never. Burnbar uses the macOS Keychain only for its own GitHub token (service ids prefixed xyz.andybowu.Burnbar). It never touches any other app's Keychain items.
Retention
If you are opted in to the public leaderboard, your daily usage rollups are retained indefinitely so your historical ranking stays intact.
If you are opted out, any usage rows are automatically purged once they are older than 90 days. A daily job removes opted-out data past that window, so opted-out usage never persists beyond about three months.
You can delete everything immediately at any time — see below.
Opt out, hide, or delete your data
You are in control from inside the Burnbar app. Nothing is uploaded until you choose to join the leaderboard.
Opt out
Turn off the leaderboard in Settings. Burnbar stops uploading and your existing data falls under the 90-day opted-out purge.
Hide
Stay opted in but mark yourself hidden to disappear from the public ranking while keeping your own history.
Delete everything
Use the delete control in Settings to wipe your account and every usage row immediately, regardless of age or opt-in status.
Contact
Questions about your data or this policy? Open an issue on GitHub or email [email protected]. Burnbar is open source — you can read the exact upload, storage, and retention code in the repository.