# Devices · Neural Inverse Govern > Device enrolment, the endpoint agent and the signed policy it enforces, device commands, wipe and lost mode, and the AI coding tools each device runs and how each is governed. Source: https://neuralinverse.com/govern/devices/ --- ![](/photos/bench.jpg) [Neural Inverse Govern / Devices](https://neuralinverse.com/govern/) # Every machine that runs AI tools, enrolled and accountable. Device enrolment, the endpoint agent and the signed policy it enforces, device commands, wipe and lost mode, and the AI coding tools each device runs and how each is governed. Endpoint agent macOS, Windows and Linux Policy Ed25519-signed, verified on the device Releases Signed by your installation's own key Lost or stolen Credential and keys revoked on the server Record Every lifecycle change audited Overview Govern manages the developer machines its endpoint agent is enrolled on, with no vendor MDM involved. Each device has its own credential and signing key, receives agent policy and agent updates signed by your installation, and reports the AI coding tools it runs. Administrators queue commands, put a device in lost mode, wipe it or offboard its person, and the server-side effects happen before the device hears about them. [What it does ↓](https://neuralinverse.com/#capabilities)[How it works ↓](https://neuralinverse.com/#workflow)[Enforced and recorded ↓](https://neuralinverse.com/#enforcement)[Where it runs ↓](https://neuralinverse.com/#where)[Get started ↓](https://neuralinverse.com/#get-started) What it does ## What device management covers today. Only what works today. Device management is still being built out; the rest is listed at the end of this page. - **Enrolment:** Enrolment tokens stored only as hashes, connect keys, and enrolment by signing in from the agent. Organisations can allow self-enrolment, with requests an administrator approves or denies. - **Device credential:** Every device call is authenticated with a server-issued device token. A device can act only as itself, in its own organisation and region. - **Signed agent policy:** The agent fetches compiled agent policy as an Ed25519-signed manifest, refuses one that does not verify or is older than the one it holds, and denies every action once its offline grace period has passed. - **Endpoint rules and kill switch:** Organisation rules and the kill switch reach endpoint agents as a signed feed, by polling and over a live connection. - **Device commands:** Refresh policy, report status, install hooks into newly detected tools, or enter lost mode, for one device or many. Only command types the agent runs can be queued, and each is acknowledged once. - **Lost, stolen and wipe:** Marking a device lost or wiping it revokes its credential, blacklists its hardware identity, flags its person's sessions of the last 24 hours for review and queues a wipe. The device counts as wiped only once it confirms. - **Unenrol:** Administrators unenrol a device with a reason, revoking its credential and signing keys. Where the organisation allows it, people can unenrol their own devices. - **Posture:** Lifecycle state, last seen, the last health report and tamper report, governed and ungoverned tool counts, and pending commands. What the device reports is labelled as reported by the device. - **AI coding tools on each device:** The tools the agent detects and how each is governed there: Govern's hook, the MCP proxy, not installed, vendor audit only or process tracking only. The server derives the mode itself. - **Ungoverned tools across the fleet:** Every AI coding tool that a live device runs without governance, with the devices that run it, most devices first. - **Tamper reports:** The agent reports tampering with itself; the device is marked tampered and each report is kept. - **Device signing keys:** Each device registers its signing key with proof of possession; keys are revoked with unenrol, wipe and offboarding. Your installation countersigns device exports and audit batches, and verifies exports on request. - **Signed agent releases:** Each installation signs every agent build with its own Ed25519 release key. Agents pin the key set at enrolment and refuse a build signed by another key, a tampered artifact and any downgrade. - **Update channels:** Per-device update channels and staged releases to a share of the fleet, with the status of each release across devices. - **Decommission and offboarding:** Decommission a device with a reason, or offboard a person and all their devices, by hand or when SCIM deprovisions them. The last Owner cannot be offboarded. - **Retention and legal holds:** Daily retention of device data per organisation, legal holds that win over it, and erasure of a person from device data with a random pseudonym. How it works ## From enrolment to retirement. 1. Request A person installs the endpoint agent and signs in, or uses an enrolment token. 2. Approve Where self-enrolment needs approval, an administrator approves the request before the device is enrolled. 3. Register The device registers its signing key with proof of possession. 4. Receive Signed agent policy, rules and releases arrive and are verified on the device. 5. Report Heartbeats carry health, tool coverage and the agent's policy decisions. 6. Retire Unenrol, wipe or offboarding revokes the credential and keys on the server. Who uses it ## Built for every role that answers for AI. - **IT and endpoint administrators:** Enrol devices, approve self-enrolment, queue commands, manage update channels and decommission machines. - **Security teams:** Mark devices lost or stolen, act on tamper reports and close the gaps in the ungoverned tools list. - **Privacy officers:** Set device data retention, place legal holds and erase a person from device data. - **Engineers:** Enrol their own machine by signing in, and unenrol it where the organisation allows. Enforced and recorded ## Decided on the server, never on the device's word. The endpoint agent's source is public, so Govern treats what a device reports as a report and acts on the server. Enforced - **Only as itself:** A device is identified from its token and can write only for itself, in its own organisation. - **Lost and wipe:** Credential and signing key revocation, the hardware blacklist and session flags happen in the request's own transaction, before the agent is told. - **Commands:** Unknown command types are refused, a command is acknowledged once, and a confirmed wipe cannot be undone by a replay. - **Releases:** The agent installs only builds signed by your installation's key, and never an older version. Recorded - **Commands:** Every queued command and every acknowledgement, recorded as the device's unverified word. - **Lifecycle:** Wipe requested and confirmed, lost, unenrolled, self-unenrolled, decommissioned and offboarded, in the change's transaction. - **Keys and releases:** Device signing key registrations and revocations, and every agent release published. - **Retention:** Every retention run and every erasure of a person from device data. Posture, health and tool coverage are the device's own report, labelled as such. No server decision relies on them. Where it runs ## Your fleet, served by your installation. [Deployment options](https://neuralinverse.com/deployment/) - **Desktop platforms:** Signed agent packages for macOS, Windows and Linux, served by your own installation. - **Self-hosted and in your cloud:** Devices enrol with, and fetch policy and updates from, your installation. - **Air-gapped:** Policy, releases and their signing keys come from your installation; the agent needs no outside service. - **Hosted by us:** We run it for you. Devices enrol in the region your organisation lives in. Get started ## Set it up with the documentation. - [Agent policy](https://neuralinverse.com/docs/govern/agent-policy/) - [Third-party AI coding tools](https://neuralinverse.com/docs/govern/third-party-tools/) - [Security architecture](https://neuralinverse.com/docs/security/) What's next ## Being built now. These are in development and not part of the product today. - **Configuration profiles:** Configuration profiles delivered to managed devices, not only stored. - **Jamf, Intune and Fleet:** Connectors that carry lifecycle actions to these MDM platforms, instead of completing them by hand. - **Neural Inverse tray:** A tray app for each person: sign in, status, the policy in force, recent decisions, each tool's control level and pause of activity capture. - **Staged device rollouts:** Canary, beta and stable rollouts of device changes that advance and roll back on their own. - **Firmware and machine identity:** Identities for hardware and firmware devices that obtain tokens with their own credentials. - **Device trust scoring:** Trust scores and fingerprint checks for devices, used in sign-in decisions. Related [Agent policy and approvals → Allow, ask and deny rules for coding agents, with named approvers.](https://neuralinverse.com/govern/agent-policy/)[Third-party AI tools → The control level Govern has over each third-party AI coding tool.](https://neuralinverse.com/govern/third-party-tools/)[Security operations → AI SOC detection, a security findings feed, incidents with two-person containment and an organisation kill switch.](https://neuralinverse.com/govern/security-operations/) ![](/photos/team.jpg) ## Talk to our team. Neural Inverse is available by request. We work with your engineering, security and procurement teams from the first conversation. [Request access](https://neuralinverse.com/contact/?topic=access)[Contact us](https://neuralinverse.com/contact/)