Skip to main content

Endpoint Agent Policy Engine

Updated on 10 November 2026.

The Endpoint Agent Policy Engine is where you decide what people can do with AI on company devices. The Endpoint Agent is the QuilrAI program installed on employees' Windows and macOS devices. It checks what people send to AI apps and what coding tools do, then applies the policies and controls you set here.

To open it, go to Govern > Policy Engine, select Endpoint Agent in the workspace selector at the top right, then select Desktop under the page title.

Go toQuilrAI ConsoleGovernPolicy EngineEndpoint Agent

Key terms​

TermMeaning
AI appAn AI chat or assistant application, such as ChatGPT, Claude or Gemini, in a browser or as a desktop app.
Coding toolAn AI coding agent on the device, such as Claude Code, Cursor or Codex CLI.
Policy and ruleA policy is a named set of rules. Each rule says who it covers, where it checks, what it looks for and what happens when it matches.
ResponseWhat happens when a rule matches: Block, Require a reason, Ask for a reason (Warn and continue in coding tools) or Record only. See Responses.
DraftYour unpublished changes to Sensitive data and Files policies. Devices never follow a draft.
VersionA numbered copy of your Sensitive data and Files policies, created by each publish, restore, Turn on, pause and resume. Devices follow the live version.
Previous Endpoint settingsThe per-app settings that devices follow while the Policy Engine is off: Guardrails, File Restrictions and Group & User Rules, in Console V1 under Settings > Endpoint. The status strip calls them existing settings. See What your previous Endpoint settings become.

What the Policy Engine controls​

  • What people share
    • Sensitive data: sensitive data in prompts, pastes and uploads to AI apps, and in coding-tool prompts, file reads and commands.
    • Files: file types and sensitivity labels in files uploaded to AI apps. These rules do not apply in coding tools.
  • What coding tools do
    • Commands & tools: shown as Not available. To limit a Sensitive data rule to certain commands or MCP (Model Context Protocol) tools, use Narrow to specific actions in that rule.
    • Skills: which skills (reusable sets of instructions) Claude Code can run.
    • Packages: whether packages that Claude Code installs are checked for known vulnerabilities. You set this in Settings > Endpoint Agent.
  • What runs and signs in
    • Apps: which applications can run on devices.
    • Accounts: which accounts and workspaces people can sign in to in AI apps.

Only Sensitive data and Files use policies. Skills, Apps and Accounts have their own editors, and Packages is set in Settings > Endpoint Agent. See Files, Skills, Packages, Apps and Accounts.

note

Policies and controls apply only to the AI apps and browsers chosen under Application Monitoring in Settings > Endpoint Agent. Older agents keep some checks for an app that is off. See Application Monitoring is off for the app.

How the page is laid out​

  1. Desktop | Browser switch. Browser holds the separate Browser Extension controls, which save at once with no draft or version. See Browser Extension policy engine.
  2. Status strip. Shows whether the Policy Engine is on, off or paused. It holds Turn on while it is off, and the More menu while it is on. See Policy Engine states.
  3. Controls tab. Its menu lists Overview (one card per area), All policies, one item per area, and Endpoint settings, which opens Settings > Endpoint Agent. All policies shows every policy and holds Priority policies, an ordered list of policies that win over other matching policies.
  4. Test tab. Shows what your live or unpublished Sensitive data and Files rules would decide, without changing anything on devices, for a request you describe (Try a situation) or for recent recorded requests (Replay activity). See Test your changes.
  5. History tab. Lists published versions to compare or restore. A restore does not change Skills, Apps, Accounts or package checks. See Review version history.
  6. Draft bar. Appears at the bottom while you have unpublished changes, with Discard and Review & publish.

Policy Engine states​

StateStatus strip showsWhat devices followWhat you can do
OffPolicy Engine off · existing settings activeYour previous Endpoint settings. Published policies are kept but do not apply.Select Turn on. You can edit the draft and use Test and History, but you cannot publish or restore.
OnPolicy Engine onYour published policies. In coding tools, file-type rules still come from your previous Endpoint settings.From More, select Pause enforcement… or Turn off Policy Engine (use previous settings)….
PausedPolicy Engine on, with Enforcement: Paused on an amber stripYour published policies, but Sensitive data and Files rules only record matches: no block, warning or reason request. Some coding-tool checks switch off. See What changes while enforcement is paused.From More, select Resume enforcement….

While the console reads the state, the strip shows Checking Policy Engine status…. If it cannot read the state, it shows Policy Engine status unavailable. Neither offers actions, and the page checks again every 30 seconds.

Turning off after you make changes

You can turn off the Policy Engine only while the running version was converted from your previous Endpoint settings alone. After anyone publishes, restores, pauses or resumes, Turn off Policy Engine (use previous settings)… stays greyed out and only QuilrAI support can turn it off. To stop blocking, use Pause enforcement. See Who can turn off.

Endpoint Agent off for the tenant

While the Endpoint Agent is off for your tenant, devices apply no policies or controls, whatever the Policy Engine state. Admins with Read Endpoint Agent see Endpoint Agent: Off for this tenant in the status strip and a banner. Turn the agent back on in Settings > Endpoint Agent.

Get started​

  1. Open Govern > Policy Engine, select Endpoint Agent, then Desktop, and read the status strip.
  2. If it reads Policy Engine off · existing settings active, select Turn on, read the dialog, then select Turn on again. Your previous Endpoint settings become policies. See Turn on the Policy Engine.
  3. In All policies, review the converted policies. Switch on a starter policy or create your own. See Policies and rules.
  4. In Test, try a few situations against your unpublished changes.
  5. On the draft bar, select Review & publish. Check the changes, wait for the safety check to pass, then select Publish changes. See Review and publish.
  6. Check the results in Findings & Interactions.

How changes go live​

ControlHow a change goes liveIn History
Sensitive data, Files, Priority policiesSaved to your draft. Goes live when it passes the safety check and you select Publish changes in Review & publish.Yes, as a new version
Skills, Apps, AccountsChanges save at once, with no draft.No
Packages and other device settingsChanges save at once in Settings > Endpoint Agent.No

Your draft exists only in your current browser tab: no other admin, tab or browser sees it, and closing the tab or signing out deletes it. See Where your draft is kept.

If a newer version goes live before you publish, you can no longer publish your draft. This happens, for example, when another admin publishes or a change to your previous Endpoint settings regenerates the converted policies. To recover your changes, see When a newer version goes live.

How changes reach devices​

ControlWhen devices apply a change
Sensitive data, FilesOn the next request: QuilrAI, not the device, checks each one against the live version.
Skills, Apps, Accounts and other controlsWhen the device next syncs with QuilrAI, within minutes. An offline device applies changes when it reconnects.

The console does not confirm that a device applied a change; check for its new activity in Findings & Interactions. For timings, see How long changes take to reach devices.

Permissions​

Without the permission it needs, a control is greyed out or hidden, with a reason such as Needs Policy update access. See Roles and permissions.

TaskPermissions needed
See policies, Test and HistoryPolicy page and Read policies
Edit the draft, publish, restore, turn on, turn off, pause or resumePolicy page and Update policies. Turn off works only while the converted version runs; otherwise only QuilrAI support can turn it off. See Policy Engine states.
See Skills, Apps, Accounts and dependency scanningEndpoint Agent page and Read Endpoint Agent
Change Skills, Apps, Accounts and dependency scanningManage Endpoint Agent

These are Access Control V2 permissions. With V1 roles, Compliance covers policies: Read to see them, Update for every policy task, including turn off. Endpoint Agent (Tenant) Read and Update cover Skills and Accounts. Changing Apps needs App Management (Tenant) Update.

Notes and limits​

  • Skills, Apps and Accounts rules, and package checks, apply whether the Policy Engine is on or off. While enforcement is paused, Skills rules, package checks and some coding-tool Accounts checks are off. See What changes while enforcement is paused.
  • Policy Engine actions never start or stop the Endpoint Agent on devices; use the Endpoint Agent switch in Settings > Endpoint Agent for that.