macOS · on-device · early access

A control tower for the AI agents on your Mac

Your Mac is quietly running AI agents that can read your files, hold your credentials, and call out to the internet. Agent Sentinel shows what each one is doing, governs what it can access, and lets you stop it — all without a byte leaving your machine.

On-device by design. No spam — just an early-access invite.

Built for Claude · ChatGPT · Grok · Kimi

The gap

We onboarded autonomous software — and never checked in

You wouldn't give a new hire root access on day one and never look again. That's effectively what we do with AI agents: they roam the filesystem, reach the network, and spawn processes — and the risk that matters isn't megahertz, it's data.

Cloud AI governance stops at the browser tab. The desktop agent on your laptop is the blind spot. Agent Sentinel closes it — locally.

Resource use
CPU · memory · processes
Network egress
which domains it reaches
Secret access
~/.ssh · keychains · .env
Audit trail
every decision, logged

See it. Govern it. Control it.

Three jobs, one menu-bar app — so the agents you rely on stay useful without going rogue.

Observe

Know what every agent is doing.

  • Live CPU, memory, network sockets, and child processes per agent
  • Rolling history that tells a brief spike apart from a runaway
  • Estimated token cost from local logs — today's spend at a glance

Govern

Least-privilege for agents.

  • Per-agent access policies for files, secrets, and network egress
  • Deny the obvious — ~/.ssh, ~/.aws, keychains, .env — by default
  • A local audit trail of every access decision. Prove it.

Control

Intervene, proportionately.

  • A real kill switch — per agent, or stop everything at once
  • Opt-in auto-stop for agents that stay critical, with guardrails
  • Quarantine and throttle on the roadmap — not just the nuclear option

100% on-device

Monitoring, policies, and the audit trail all stay on your Mac. No cloud, no data leaves the machine. For a tool whose job is trust, that's the only posture that makes sense.

Report-only first

Governance starts by showing what a policy would block — so you see the value before anything is enforced. Enforcement is opt-in and guarded.

Native & open core

A native SwiftUI menu-bar app with an open-source core. Lightweight, inspectable, and honest about what it can and can't do.

Get early access

Agent Sentinel is in active development. Join the waitlist and we'll invite you to the first cohort — and we'd love to hear what you most want to see or control about the AI agents on your machine.

On-device by design. No spam — just an early-access invite.

Questions? hello@promptshields.com