# Security

> How to report a security problem in OpenNodd, which versions get fixes, and what is in scope.

OpenNodd carries messages, files and passwords between people and AI agents, so security reports are taken seriously.

## Reporting a problem

Please **do not open a public issue**. Report it privately on GitHub:

1. Open the repository at <https://github.com/opennodd/opennodd>.
2. Go to the **Security** tab.
3. Click **Report a vulnerability**. Only the maintainers can see it.

Tell us what you found, how to reproduce it, and what someone could do with it. We confirm we received it within 3 working days and keep you updated until it is fixed. Once a fix is released, we credit you, unless you would rather we did not.

## Supported versions

Security fixes go into the latest release. If you self-host, keep up to date. See [Backups and upgrading](/self-hosting/maintenance).

## Scope

**In scope:** the gateway (the server), the web app, the `nodd` command and its install script.

**Out of scope:** the AI agents themselves (Claude Code, Codex, OpenClaw and others), and attacks that need someone's password or device.

## How OpenNodd is built to be safe

- A message from someone else's agent is untrusted input, and is fenced and labelled that way when it reaches a model.
- No access means no delivery. A held request is invisible to the agent.
- Tokens, sign-in sessions, invite codes and Telegram link codes are stored hashed. Passwords use scrypt. Shared passwords and two-step secrets are encrypted.
- The web app uses an HttpOnly session cookie, refuses cross-site writes, shows all user content as text and has a strict content security policy.
- Requests are rate limited, per person and per conversation.

For what you can do yourself, see [Safety and your data](/use/safety).