Security

How we protect your books.

Your books are the most sensitive thing about your business. Here is what we actually do about that, in plain terms.

Your login

  • Passwords are stored as Argon2 hashes. We never store or see your actual password.
  • Your session uses a short-lived access token held only in memory and a rotating refresh token in a cookie that page scripts can't read. If an old token is ever replayed, that whole session is signed out.
  • Sign-in and sign-up attempts are rate-limited to slow down guessing.

Your data stays yours

  • Each business's records are separated by the database itself (row-level security), not just by application code. A bug in our code can't quietly show one business another's numbers.
  • All traffic uses HTTPS, and the app sends strict browser security headers, including a nonce-based content security policy.
  • We never ask for your bank login. You bring a statement; nothing is connected.

AI and your documents

  • When you upload a receipt or a PDF statement, or ask the assistant a question, that content is sent to Anthropic's Claude API to be read and answered. CSV imports never leave our servers for AI.
  • The model can only read your books through a small set of read-only questions. It cannot post, change or delete anything. It can draft a suggestion, such as filing several charges under one category, but nothing happens until you press Apply, and every applied item can be undone.
  • We don't trust what it returns: amounts are re-checked with plain arithmetic, statements must add up to their own closing balance, and every suggestion waits for your approval.
  • Your original files are stored with your books so every number can be traced back to its receipt.

Books that can't be quietly rewritten

  • The ledger is append-only. Posted entries can't be edited or deleted; a mistake is corrected by a visible reversal.
  • Every entry must balance, enforced by the database.
  • An audit trail records who changed what, when, and what it looked like before.

What we can't claim yet

  • LedgerAI is early. It hasn't had an independent security audit or a compliance certification yet.
  • Backups and point-in-time recovery are being upgraded as we grow.

Found something that looks wrong? We want to hear about it. Meanwhile, you can set up your books and judge for yourself.