Skip to main content
Privacy and Data cover — agent activity on the left, your life on the right. AgenShield is installed on machines people work on, so it is fair to ask exactly what it watches and what leaves the device. This page answers that. If you are evaluating AgenShield for a fleet, this is also the page to share with your privacy or works-council reviewers.

​
What AgenShield observes

AgenShield’s scope is AI agents and the processes they start — the execution tree behind each agent, as defined by your organization’s policy. It records what those agents execute, which files they read and write, and which destinations they connect to. Anything unrelated to your AI agents is not observed at all — programs and tools an agent starts are within its observed scope, everything else on the Mac is not. agenshield status lists the agents AgenShield has detected on this Mac. See How AgenShield works. AgenShield does not record keystrokes, capture the screen, or browse your files on its own. A personal file appears in its records only when an AI agent reads or writes it — that access is the event being recorded. It has no view into applications unrelated to your AI agents.

​
What leaves the device

Activity records are sent to your organization’s AgenShield backend and are visible in the Frontegg Portal. Records are signed by the device and sent over TLS. Your administrator chooses how much detail is included; the maximum available is:
  • which agent acted, and what it did (run, read, write, connect)
  • the file paths and destinations involved
  • the policy decision — allowed, blocked, or observed
  • device health: version, extension status, last policy sync
Administrators can reduce this. A common configuration keeps the signal — “this agent read a credentials file”, “this agent reached an unapproved host” — while dropping the literal paths and destinations.

​
Secrets are removed before anything is sent

Anything that looks like a credential is detected and stripped on the device, before the record is transmitted. API keys, tokens, passwords, and private keys are replaced with a marker naming the kind of secret found. The secret value itself is never sent and is never written to a log. This runs on the device rather than in the cloud on purpose: a secret that leaves the machine has already leaked, no matter what happens to it afterwards.

​
Network inspection

If your administrator enables traffic inspection, AgenShield can look inside an AI agent’s encrypted connections to record what the agent actually sent. This requires a certificate to be installed on the Mac and is off by default. When it is on:
  • it applies only to AI agents, never to your browser or other applications
  • secret redaction runs on inspected content, same as everywhere else
  • the certificate is generated per device and its private key never leaves it
See Network inspection certificate for what the certificate is and how to verify it.

​
What your administrator can see

​
Local data

Activity is also stored on the device so protection continues while offline, and so diagnostics are available. Local storage is capped and old records are removed automatically. When AgenShield is uninstalled, local data is removed with it.

​
Diagnostics you send manually

A diagnostics bundle is generated locally and goes nowhere until you send it. Review it before sharing — see Collecting diagnostics, which lists what the bundle contains and what it never contains.