Security model in depth
The AI never gets a shell account. It gets tools whose limits are enforced on the device and whose risky calls a person approves. This page collects the layers in one place; the short version is on the Security page.
On the device
- No listening port: the agent connects out over TLS. No SSH key on the machine is needed for it.
- Least privilege: a normal user by default (wecommander on Linux, your own user on Windows and macOS); root only with two keys, Linux only.
- Allowed folders: paths are resolved before they are checked, and symbolic links leading outside are refused.
- Local switches (commands off, file writes off, root ceiling, redaction, system logs) live only in the agent's config; the gateway and the AI cannot change them.
- High-risk actions run only with an approval signed for that device and that exact action; the master signing secret never leaves the gateway, each device gets its own derived key.
On the gateway
- Role checks: viewers read, operators run tools, admins and owners approve and manage.
- Approvals bound to the exact action, single use, with expiry.
- The safety net and the allowed-command rules, re-checked on every match.
- AI clients and API keys cannot change permissions, allowed commands or autonomous mode, and cannot approve.
- Audit log with secrets masked; request and response sizes capped; rate limits on sign-in, pairing and invitations.
Accounts and sign-in
- Passwords are stored as salted scrypt hashes; session cookies are HttpOnly, SameSite=Strict and Secure, with CSRF protection.
- MCP clients use OAuth 2.1 with PKCE: one-time codes valid 5 minutes, access tokens for 1 hour, rotating refresh tokens for 30 days. OAuth tokens work only on /mcp.
- Disabling a user or removing them from the workspace ends their access at the next request.
- Device credentials and pairing codes are random and stored only as hashes; pairing codes are single use and expire.
Your data
- SSH passwords and keys stay in your computer's keychain; sync and the team library are end-to-end encrypted, and terminal contents do not pass through WECOMMANDER's servers.
- With Hybrid Access the server's address, user and credentials never leave your computer.
- Card details are handled by Paddle and never reach WECOMMANDER.
Honest limits
- Commands run with the rights of the user they run as; a command can read whatever that user can, even outside the allowed folders. Grant the agent's user access only to what the AI should use.
- Over SSH there is no folder limit.
- Secret masking and the safety net are guards, not guarantees.
- Security model: How WECOMMANDER keeps AI access to your machines under control: outbound-only agent, allowed folders, approvals, secret masking, audit log.
- AI permissions: trust levels, allowed commands, root and the safety net: Set per device and per SSH server what AI may do without asking: strict, standard or full trust (time-limited or not), allowed command patterns, run as root with two keys, the safety net and autonomous mode.
- Secret masking: How WECOMMANDER masks passwords, tokens and keys in command, terminal and log output on the device before the result reaches the AI, and what is not masked.
- Hybrid Access: AI on SSH servers without an agent: Share saved SSH servers with AI clients through WECOMMANDER on your own computer (Windows, macOS, Linux): nothing installed on the server, keys never leave your computer, approvals and AI permissions apply.