# Security

URL: https://docs.vaultpanel.net/guides/security

> How VaultPanel protects the panel and your servers, and where to manage firewalls, SSH, brute-force protection, updates, sign-in and access.

VaultPanel manages servers as root, so its design keeps that power narrow: the panel itself runs without root, every change is a signed job from a fixed allow-list, and every action is audited. This section covers the security model and the settings you manage, on each server and for the panel.

## Security overview

**Security > Security Overview** scores every server on what VaultPanel checks from the summary each agent reports every 5 minutes:

| Check | OK when |
| --- | --- |
| **Firewall** | UFW is installed and on, and does not let in everything no rule blocks |
| **Root sign-in over SSH** | Root cannot sign in with a password (keys only, or not at all) |
| **SSH passwords** | Only SSH keys are accepted |
| **Brute-force protection** | Fail2ban is running and watches SSH |
| **Automatic security updates** | Security updates install every day |
| **Waiting security updates** | None are waiting |
| **Agent privileges** | The agent runs as its own user, with a root helper (agent 0.36.0 and newer) |

**Restart needed** appears as information when installed updates need a reboot. The overview also counts servers with the firewall off and agents too old to report security settings (firewall, SSH, Fail2ban and update management need agent 0.6.0 or newer).

The same checks, and every setting behind them, are on each server's **Security** tab. Reading them needs `security.read`; changing them needs `security.manage`. Both belong to owners and admins by default, and every change runs as a job and is recorded in the audit log as `security.*`.

## How the panel is protected

- **No root.** The API and the worker run as the unprivileged `vaultpanel` user in a systemd sandbox: no capabilities, `NoNewPrivileges`, a read-only system with write access only to the panel's own folder, and private `/tmp` and devices. `.env` and the TLS key belong to root and are only readable by the panel. The single root step connects the panel's own server before the API starts and takes no input from the network.
- **HTTPS from the first start.** A new install serves the panel over HTTPS with a self-signed certificate, so passwords, the setup token and session cookies never cross the network in clear. Plain HTTP is available only on `127.0.0.1`. HSTS is sent only where the certificate is trusted, on a [panel domain](https://docs.vaultpanel.net/getting-started/panel-https-and-domain).
- **Sessions and CSRF.** Session tokens are random 256-bit values stored hashed, in `HttpOnly`, `SameSite=Strict` cookies that are `Secure` in production, lasting 12 hours (`VAULTPANEL_SESSION_TTL_HOURS`). State-changing browser requests also need a per-session CSRF token. Changing a password, role or active state ends sessions.
- **Rate limits.** Sign-in, setup, password reset and agent registration are rate-limited per IP address, and the whole API allows 600 requests per minute per address. See [Sign-in and two-factor](https://docs.vaultpanel.net/guides/security/sign-in-and-two-factor#rate-limits-and-lockout).
- **Validated input.** Every route validates its input, database queries are parameterized, CORS is closed by default and errors never include stack traces; a request ID links them to the server log.
- **Secrets at rest.** Passwords are scrypt hashes; session, enrollment and reset tokens are stored as SHA-256 hashes. Two-factor secrets, the job-signing key, app and container environment variables, database passwords, webhook URLs and secrets, and AI provider keys are encrypted with AES-256-GCM using keys derived from `VAULTPANEL_SECRET_KEY`.
- **Outgoing calls.** Webhooks and plugins never call loopback, link-local (cloud metadata) or reserved addresses, and reach private networks only with `VAULTPANEL_WEBHOOK_ALLOW_PRIVATE=true` (off by default). The host is resolved before each attempt and the request goes to the checked address, so DNS cannot be switched after the check.

## How servers are protected

- **Signed, allow-listed jobs.** The agent signs every request it sends, and the panel signs every job. The agent runs only operations in its allow-list, re-validates their parameters, and runs fixed commands without a shell. There is no endpoint that runs arbitrary commands.
- **A root helper, not a root agent.** From agent 0.36.0 the agent that talks to the panel runs as its own user with no capabilities. A separate root helper runs only jobs and terminal sessions it has checked against the panel's signature itself. See [Connecting servers](https://docs.vaultpanel.net/guides/servers/connecting-servers#how-the-agent-runs).
- **Local policy wins.** Each server's `config.json` can refuse whole areas, such as reboots, firewall changes or the terminal, whatever the panel asks.
- **Lockout protection.** The agent refuses firewall rules and SSH changes that would close SSH or the panel, or leave no administrator able to sign in. Such requests fail with `LOCKOUT_REFUSED`.
- **Isolation.** Each website and app runs as its own Linux user. App units also get `ProtectHome`, kernel and control-group protections, and optional CPU and memory caps.

## Managing security

- [Firewall](https://docs.vaultpanel.net/guides/security/firewall)
- [SSH](https://docs.vaultpanel.net/guides/security/ssh)
- [Fail2ban and updates](https://docs.vaultpanel.net/guides/security/fail2ban-and-updates)
- [Users and roles](https://docs.vaultpanel.net/guides/security/users-and-roles)
- [Sign-in and two-factor](https://docs.vaultpanel.net/guides/security/sign-in-and-two-factor)
- [Panel access by address](https://docs.vaultpanel.net/guides/security/panel-access)
- [Audit log](https://docs.vaultpanel.net/guides/security/audit-log)
- [Best practices](https://docs.vaultpanel.net/best-practices)

## Reporting a vulnerability

Do not open a public issue. Report it privately through GitHub: **Security > Report a vulnerability** on [VaultPanel/packages](https://github.com/VaultPanel/packages), VaultPanel's public repository. Include what an attacker can do and what they need first, the steps to reproduce it, and the VaultPanel and agent versions (`vaultpanel version`).

You get an acknowledgement within 3 working days, and an assessment within 10 working days. Security fixes go into the latest release, so keep the panel updated (`sudo vaultpanel update`) and update agents from the panel.
