Trust rules

Nobody should run an unknown program on a production server unless these rules hold. They are part of the product, and no feature is worth breaking one.

Rules for the agent

1. Read-only

It never changes a file, a setting or a container. It prints the fix, and you run it.

2. Open source

The agent’s code is published at release, so anyone can read what it does before installing it. Until then, what the agent reads and sends lists every file it reads and everything it sends.

3. Local by default

A one-off scan sends nothing off the server. Data leaves only after you connect the server, which a later release adds.

4. Secrets never leave

The agent reports that a secret exists and where, never its value. Passwords in logs and connection settings are masked on the server.

5. Outgoing only

The agent opens no port and accepts no incoming connection.

6. Easy to remove

One command uninstalls it. One click deletes a server’s history.

7. No AI and no third parties

Logs, errors and findings go only to the service’s own storage.

Rules for the service

These hold for the hosted dashboard when it arrives.

1. Two-step sign-in is required for every teammate, in the dashboard and in the terminal.
2. Each token does one job. A server token can only send results. A deploy token can only mark deploys. Neither can read data.
3. Every action is recorded in the activity log: who assigned, muted, invited or changed what.
4. Every request is checked against the person’s role and access, on the server. Anything hidden from them answers as not found.

See for yourself

What the agent reads and sends lists every file and setting a scan reads. It is built from the same file the agent uses.