Account security
Two-factor authentication, recovery codes, passwords and SSH keys — and what each one protects.
Your account is the perimeter. Every project you belong to, every workload you can stop and every shell you can open is reachable from it.
Two-factor authentication
Turn it on from your account settings. It is an authenticator app — a time-based code, the standard kind, so any TOTP app works. There is no SMS option, which is deliberate: SMS is the weakest common second factor.
Enrolment shows a QR code and the secret behind it. Scan it, then confirm with a live code — enrolment does not complete until the platform has seen your app produce a working code, so you cannot lock yourself out by mis-scanning.
The secret is encrypted at rest with the same key that protects cluster credentials.
Recovery codes
Confirming enrolment gives you a set of one-time recovery codes. They are shown once.
Save them somewhere that is not the phone holding your authenticator, because their entire purpose is the day that phone is lost. Each works once.
You can regenerate the set at any time — which invalidates the old one, so any code you wrote down before stops working. Regenerating requires a current code, as does turning two-factor off. Someone with your password alone can do neither.
Passwords
Change yours from account settings.
If you have forgotten it, the reset flow emails a link. Should mail not be configured on your platform, an operator can issue a reset link directly — ask rather than waiting for an email that is not coming.
Passwords are stored as Argon2 hashes. Nobody, including your operator, can read yours.
SSH keys
Adding a public key to your profile lets you open a shell on your workloads from your own terminal. The key identifies you, so it inherits your project roles — it is not a separate grant, and it reaches only workloads you could already reach. See Logs and shell access.
Remove keys you no longer use. The console shows each key’s fingerprint so you
can match them against ssh-keygen -lf rather than guessing from a name.
API keys are credentials too
An API key is a long-lived credential that acts as you. Give each one an expiry, pin it to a project when you can, and check the last-used column occasionally — an unused key is a key nobody would miss if it leaked.
What to do if something is wrong
Revoke first, investigate second. API keys, inference keys and SSH keys all revoke immediately, and revoking is reversible in the sense that you can always mint a new one.
Then read your project’s audit log. It records which credential performed each action, so “was it me” is a question with an answer.