AgentLayer▸docs
Run it

Privacy

All data stays local. Secrets are deny-gated. Transcripts are redacted before they persist.

  • All data stays local. The home is markdown on your disk. No cloud sync unless you commit it somewhere.
  • Secrets live in the deny-gated .state/secrets/.env: API keys, database connection strings, Google OAuth. Non-secret config lives in .claude/settings.local.json, runtime state in .state/. All gitignored by default, and the agent's own shell is blocked from reading the secrets store on both hosts. Which key goes in which file: Configuration.
  • Transcripts are redacted before persisting. Session capture masks every value it finds from the secrets store, and anything shaped like a common key format. Redaction has the passes.
  • Nothing on the dashboard leaves your machine. The page makes zero external requests: no CDN, no webfonts, no analytics.
  • Keys never enter chat. Neither init nor configure-skills asks for a credential in conversation, because a pasted secret would land in the transcript and the model API. You fill values in your editor.
  • Telemetry, crash reports and transcript uploads are off. On Claude Code, the home's .claude/settings.json sets DISABLE_TELEMETRY, DISABLE_ERROR_REPORTING, DISABLE_FEEDBACK_COMMAND (no /feedback, /bug or /share, which send the transcript) and CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY. The home doesn't set CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, because it also hides /design and artifacts and blocks Remote Control. On Codex, the user-level note the plugin generates carries the analytics and telemetry keys.
  • Turn off model training in your Claude account. On Free, Pro and Max, Anthropic trains on chats and Claude Code sessions unless you opt out, and keeps them up to 5 years instead of 30 days. In claude.ai or the mobile app: your name → Settings → Privacy → Help improve Claude → off. Team, Enterprise and API accounts aren't trained on by default. See Anthropic's privacy article.
  • Outside text stays marked untrusted. Ticket bodies, reviews, pull-request diffs and search results reach the agent inside an untrusted-content wrapper. A wrapper tag spelled inside that text is defused, so the source can't close the wrapper early and speak outside it, and an attachment read from its temp folder counts as untrusted too.
  • The folder is the product. If anything goes wrong, the markdown plus its git history is the complete backup.

Keep homes private

dashboard.html sits at the home root and reflects your tasks, knowledge stats, and redacted settings. If you publish the home repo, it publishes too.