releavo

Security

How Releavo keeps your projects safe, and how to tell us about a problem you've found.

How Releavo protects your work

  • Permissions live in the database. Every read and write from the app runs as the signed-in person, and row-level security decides what they can reach. Approving a doc is checked by the database too, not only the screen.
  • The AI can't approve. Releavo's agent and connected AI apps write drafts; the person whose role owns a doc approves it.
  • Sign-ins stay with us. Connector credentials are stored encrypted and never reach the AI. Apps connected over MCP get Releavo's own tokens, which only work on Releavo's MCP server, never the person's sign-in.
  • Code goes to its own branch. When Releavo writes code, it works in an isolated dev box and pushes only to its own releavo/ branch, as a draft pull request someone reviews.

Report a problem

Email [email protected] with "Security" in the subject. Please include what you found, the steps to see it, and what an attacker could do with it. We'll reply within five working days and keep you posted until it's fixed.

While you look:

  • Test only with accounts and projects you made yourself. Don't read, change or delete anyone else's data, and stop as soon as you've shown the problem.
  • Don't run denial-of-service or high-volume tests, and don't test people (phishing, social engineering) or our providers.
  • Give us a reasonable time to fix it before telling anyone else.

If you follow these, we won't take action against you for your research, and we'll credit you when it's fixed if you'd like.

We don't run a paid bug bounty.

What we already know about

These aren't problems on their own; we'll look at them when you show real harm they lead to:

  • Anyone can register an MCP app at /oauth/register. MCP apps register themselves, so this is how it's meant to work. A registered app gets nothing until a person signs in and allows it, and an app on a website Releavo doesn't know needs the person to confirm that site.
  • Reports from automated scanners without a working attack: missing headers on pages that hold nothing private, version numbers, rate limits, email records (SPF, DMARC), clickjacking on pages with no actions.