EverShell

Security

Built to sit on the internet.

A browser terminal for your servers is a powerful thing to put online. Here is what protects it, and what it cannot protect you from.

Start with the honest part.

Anyone who gets all the way past the login gets, on every host their account reaches, the access of the user its agent runs as, much like holding an SSH key for that user. Roles limit which hosts an account reaches and whether it gets a shell there, not what a shell can do. So the work goes into making “all the way past the login” hard: below is what a stolen password alone does not get you.

That is why the defaults are strict: HTTPS on a real domain, passkeys, and an agent that runs as a dedicated, unprivileged webterm user unless you decide otherwise.

Read the full threat model →

A stolen password is not enough.

  • Accounts with 2FA

    Signing in also asks for the authenticator code, or is done with a passkey. A new passkey cannot be added without that second factor, so the password cannot be turned into a key of its own.

  • Hosts marked “Require 2FA”

    Every session, file action, command or update there needs a fresh passkey, authenticator code or SSO re-authentication; the account password alone is refused. Automation tokens are refused outright.

  • The window stays short

    After the second factor, the host stays open for 5 idle minutes and at most 60 minutes in all; then the terminal locks, shared viewers included, until you present a factor again.

  • What it cannot cover

    An account without any second factor, and hosts without the flag, are protected by the password alone; the Security card on the dashboard points out both. A device that is already signed in and inside its window is, for that window, you.

Getting in

  • Passkeys

    WebAuthn sign-in, with TOTP and recovery codes as alternatives.

  • Step-up per host

    Flag a host and everything there asks for a fresh second factor, never the password alone.

  • Single sign-on

    OIDC (Authentik bundled or your own) and SCIM provisioning; group-to-role mapping.

  • Lockout and alerts

    Failed sign-ins are throttled; new sign-ins and security changes are emailed.

After the login

  • Roles and scopes

    Owner, Admin, Operator, Viewer or your own roles; per host or folder; time-bound grants; crown-jewel hosts.

  • Command guardrail

    Admin rules that block a command, or ask for confirmation, before it runs, including in split views and Run on hosts.

  • Shares that cannot outgrow you

    A share or replay link never gives more than its creator still has, and dies when that access ends.

  • Audit log

    Who did what, from which IP. Tokens and secrets are kept out of it.

  • Security card

    The dashboard tells you what needs attention: a missing passkey, an open share, an old agent.

  • Scoped automation tokens

    Read or run only, expiring, capped by a role and a host scope, refused on 2FA hosts.

On your hosts

  • Nothing to expose

    The agent dials out. No port opens on the server, and the gateway stores no login for it.

  • One token per machine

    Bound to that host and individually revocable. Reinstalling an agent issues a new one.

  • A client certificate per agent

    From 3.7.0, each agent also proves itself with a certificate from your installation's own authority, checked during the TLS handshake. A client without one never reaches the gateway, and a copied token alone opens nothing.

  • Signed updates

    An agent only runs an update signed with your own key.

  • SSH host keys pinned

    A changed key raises an alarm instead of connecting.

Running it

  • Encrypted backups

    Daily, encrypted, and optionally off-host: Google Drive, Dropbox, SFTP or FTPS.

  • Pinned releases

    The upgrade script takes a version tag, not latest, and pins it to an immutable image digest before it switches.

  • Release checks

    The publish pipeline requires unit, browser and accessibility tests, a secret scan and a Python dependency audit. The image and npm scans run weekly, separately.

  • Reporting a vulnerability

    Privately, through GitHub. See the security policy.