Privacy model
Your code and prompts stay local. Only usage metadata leaves your device — and only when you turn sync on.
Never collected by default
- Prompt text and responses
- Source code and file contents
- Terminal output
- Secrets and API keys
- Full file paths
What metadata we collect (when you opt in)
- Provider, platform, and model
- A repository identifier and branch (hashed)
- Session ID, timestamps, and duration
- Token counts, cache reads/writes, reasoning tokens where exposed
- Provider-reported and API-equivalent cost
- Tool-call counts, files-changed count, commits
- An optional task label you choose
Tracker modes
- Local — nothing leaves your machine.
- Private sync — metadata goes to your private dashboard only.
- Published — you choose which aggregate data appears publicly.
Browser extension
The browser extension estimates visible chat text locally. By default it never syncs raw messages, page HTML, attachments, or your identifiers from the AI platform. It never claims an exact provider token count when the provider does not expose one.
Uploads
Any evidence you upload is routed to a private quarantine, scanned and redacted for secrets, and — if you chose metadata-only retention — the raw file is deleted after metadata extraction.