The platform security model

A platform that can open a shell on every machine you manage is a high-value target, and it is more useful to describe its security model plainly than to gesture at it.

Operator authentication is email and password with optional TOTP multi-factor. Login attempts are recorded whether or not they succeed. There is no SAML or OIDC federation in the product today, so operator identity does not flow from your identity provider — enable MFA on every account and manage the operator list actively, because those are the controls you actually have.

The agent connectivity model is outbound-only. Endpoints establish connections to the platform; the platform never connects inward. That removes an entire class of exposure — there are no inbound ports to leave open, no management interface reachable from a network you do not control, and no requirement to expose endpoints to reach them.

Remote sessions require a short-lived ticket issued to an authenticated operator, and every session is bound to that operator identity and recorded. Privileged actions across the platform are written to an append-only audit log carrying actor, action and target.

The credential vault holds shared operational credentials with an explicit reveal action, and reveals are recorded. The design goal is that access to a client credential is a deliberate act with a name attached rather than an ambient property of being logged in.

Responsibility is shared, and the boundary is worth being explicit about. The platform provides authentication, isolation, auditability and a defensible connectivity posture. Whether operator accounts are individually named, whether MFA is enrolled, whether leavers are revoked promptly, and who is allowed to open a terminal on a production server are all your decisions. Most real compromises of tools in this category are not exploits; they are an unused account with a reused password and no second factor.