AgenShield

Endpoint security for AI coding agents on macOS — see what your agents actually do, then govern it with centrally managed policy, without changing how developers work.

AI coding agents run commands, read files, and reach the internet on behalf of a developer — at machine speed, often without anyone watching. AgenShield puts a boundary around them: it finds every agent, applies policy your security team defines centrally to what each may run, read, and reach, and records every decision.

Developers keep using Claude Code, Cursor, or Codex the way they already do.

The idea in one page

1. AgenShield finds the AI agents on the Mac

   Every installed agent — Claude Code, Cursor, Codex CLI, Gemini CLI, and many more — is detected automatically and governed as its own actor.

2. Policy arrives from your organization

   The Mac receives a signed policy from the Frontegg Portal and applies it locally — including while offline.

3. Every action is evaluated against that policy

   AgenShield follows each agent’s execution tree — the agent, the tools it runs, and their subprocesses — and checks everything they run, read, and connect to. No per-agent setup, no new commands for developers.

4. You start by watching, not blocking

   In monitor mode nothing is blocked; activity is simply recorded. That evidence is what tells you which rules are worth enforcing.

5. You promote rules to enforcement one at a time

   Turn on blocking for a specific rule once you have seen the real traffic — never fleet-wide on day one.

That loop — observe, decide, enforce, repeat — is the whole product.

Where to go next

How AgenShield works

The model behind the loop above: policy-governed agents, the execution tree, the three enforcement modes, and what AgenShield deliberately never touches. Read this before deploying.

Quickstart

Take one Mac from nothing to governed agents in about ten minutes.

What gets installed

Every component that lands on the Mac, what each is responsible for, where it lives, and how to confirm it is healthy.

Deploy to a fleet

Zero-touch rollout through Jamf, Kandji, Mosyle, JumpCloud, or Intune — no prompts on the endpoint.

Rollout playbook

The recommended path from first pilot Mac to enforced policy, and how to avoid breaking developer workflows on the way.

Working with your agents

For developers: what changes day to day, and what to do when something is blocked.

The AgenShield app

The menubar icon and the dashboard — status, activity, the policy in force, and one-click diagnostics.

The Frontegg Portal workflow, end to end

If you are the person setting AgenShield up, this is the order to do it in — all of it in the Frontegg Portal. Each page hands off to the next.

#	Step	You end up with
1	Create an install campaign	An enrollment token and ready-made deployment artifacts
2	Enroll devices	Macs reporting in, with no prompts on a managed fleet
3	Read the fleet	Health, coverage, and which policy revision is actually live
4	Watch telemetry	Evidence of what your agents really do
5	Write rules	Narrow, agent-scoped policy built on that evidence
6	Set the enforcement mode	Monitor first, then one rule at a time
7	Govern agent resources	Reviewed skills and connectors instead of whatever appeared

Who each section is for

You are…	Start with
Evaluating AgenShield	How AgenShield works, then Privacy and data handling
Rolling it out to a fleet	Install campaigns, then MDM deployment
Planning the rollout sequence	Rollout playbook
Setting up your own Mac	Quickstart
A developer whose agents are governed	Working with your agents
Supporting someone else	Common issues

Good to know up front

• Policy is managed centrally. Developers cannot change policy or turn enforcement off locally — changes go through the Frontegg Portal. (The one local exception is the administrator-only emergency recovery below.)
• Protection survives being offline. The last policy the Mac received stays in force, and activity is recorded locally and uploaded when connectivity returns.
• It cannot lock you out of your Mac. The limits that protect your own account and macOS itself are built into the signed security extension and are not configurable — not by policy, not by an administrator, not by anyone with access to the device. If AgenShield ever misfires it heals itself automatically, and an administrator can always emergency-disable enforcement locally.
• macOS 14 or later, Apple silicon. Intel Macs are not supported yet. Intel and Linux support are in development. Windows has a one-line install today (see Install and uninstall), but enforcement is not yet at full parity with macOS — see Windows install warnings for two known limitations (an unsigned installer that triggers a Windows SmartScreen warning, and the command-line enrollment token method — not the install link or a Group Policy/Intune push — not completing on non-English Windows editions).