Managed devices

Optional. Agents reach Tier Two with no install at all; the Client is for organizations that want the machine to count as well as the person.

What it adds

  • A device identity: a key the machine generates locally and never exports.
  • Posture reporting, so a stale or unmanaged machine is visible rather than assumed fine.
  • Automatic registration of the endpoint in the MCP hosts on that machine.
  • Fleet enrollment: push it by MDM or bake in an admin-minted enroll secret.

Download the Client (v0.3.1) from /download. macOS packages are signed with a Developer ID certificate and notarized. There is no Windows build yet.

The trust ladder

LevelWhat it takes
BasicThe floor: the person signed in. No claim at all is made about the machine.
ManagedThe device was enrolled by IT (an admin-minted enroll secret, MDM, or explicit admin attestation) and a Client on it has reported posture recently.
Verified ownerManaged, plus the owning person is identity-proofed. Cumulative: identity proofing never substitutes for machine governance.
Managed is earned, never self-declared. A call from a machine with no Client is Basic, and a tool that requires more refuses it: closed, not open.

Enrolling a fleet

Machines enroll without a human present and land unassigned: they are inventory, not authorization. No governed call from an unassigned device succeeds until someone binds it to its owner at first sign-in. Revoking a device stops it, and every enrollment, binding and revocation is a row in the governance ledger.

Posture is reported by the Client on that machine and is not independently verified. A compromised host can lie about itself. Treat it as a signal for policy, not as proof.

Device health

Every machine running the Client shows as Healthy, Unhealthy or Unknownunder Agents → Devices, judged against the checks a workspace admin requires (Settings → Advanced → Device health). Health is derived from the Client's own posture report, signed with that machine's key; it is never stored, so what the page shows is what the endpoint would decide.

CheckWhere it comes from
Disk encryptionFileVault on macOS, an active dm-crypt device on Linux.
Screen lockThe lock-on-wake setting on macOS, the GNOME screensaver lock on Linux.
FirewallThe macOS application firewall; firewalld or ufw on Linux.
Secure bootThe UEFI variable on Linux (not reportable on macOS).
System Integrity Protection, GatekeepermacOS only.
Minimum OS versionThe product version against a per-platform floor you set.
Client up to dateThe Client's version against the newest release.
Recent posture reportHow old the newest report may be.
  • Only a required check that affirmatively fails makes a device Unhealthy. A check the Client cannot report is Unknown and never counts against it.
  • Nothing is refused until an admin turns on Enforce. With Enforce on, an Unhealthy machine loses Managed, and tools that require a managed device refuse it with a reason that names the failing checks. Tools with no device requirement are never affected.
  • The Client re-checks every hour, on wake from sleep, on unlock, and on demand from its tray; the tray shows the same verdict the dashboard shows.
Posture is a self-report. A compromised host can lie about itself, which is why enforcement is a choice you make per workspace, and why the dashboard labels it as such everywhere it appears.

Revoking a device

Revoking a machine revokes the device and every agent seen on it in the last 30 days, and closes their live sessions. A call that still carries the machine's device identity is refused outright from then on, on any tool, and lands in Activity as device revoked rather than quietly dropping to Basic. The confirmation names the agents it will take with it.

A hosted agent is one per person and host app, so revoking it signs that person out of that host everywhere until it is restored: by the person from Settings → Agent access, or by an admin from the agent's page. Revoking is a kick; removing someone from the workspace is the ban.