Privacy

Privacy Policy

Atlas is local-first. Repository indexing happens on your machine. Atlas does not upload repository contents during indexing. Atlas never trains AI models using customer repositories.

What runs locally

  • Repository scanning and file reading
  • The dependency graph, subsystem map and architecture analysis
  • Risk detection, impact analysis and Plan Change workflows
  • Generation of the AI context packet

Your source code is read and analyzed entirely on your device. Atlas does not upload your repository to an Atlas-hosted service.

What can leave your device

  • The context you copy or your agent requests. When you click “Copy for Claude/Codex/Cursor”, the compact packet goes wherever you paste it. If you connect an external coding agent over MCP, that agent may send the context you request to the model provider it is configured to use — the same as when you use that agent without Atlas.
  • Account & licensing (if you sign in): email and license status are exchanged with the Atlas accounts service to validate your plan. Your code is never included.
  • Pseudonymous product analytics to improve Atlas (can be disabled in Settings). Events carry a random per-installation identifier — and your account id if you are signed in — but never repository contents, repository names or paths, file names, source code, prompts, answers, secrets, or MCP payloads.
  • Billing metadata. Paid subscriptions are handled by Paddle as Merchant of Record. Atlas does not store your payment details.

Trust by construction

Atlas refuses to hand your AI stale or unverifiable context. If a repository changed since the last scan, exports are blocked until you refresh — so your agent never works from a misleading map.

Data & contact

For privacy questions, data requests, or security disclosures, contact us via the yoavkozlovski@gmail.com.

Last updated: 2026-07-09.