Features

How Does a Windows AI Security Agent Work?

A Windows AI security agent works by running as a background service on the device and inspecting outbound traffic to known AI services before it leaves the machine — no browser extension, no manual proxy setup.

How SafePrompt does it

SafePrompt installs an MSI package containing a background Windows service, SafePromptWatchdog, registered with the Windows Service Control Manager. It starts automatically, survives reboots, and keeps the agent's detection engine and policy up to date.

When a user sends a prompt to ChatGPT, Claude, Gemini, or Copilot, the agent inspects it on the device. If it matches a policy — a password, an API key, personal data, or a custom keyword rule — the agent can warn the user, redact the sensitive part, or block the request outright, all before it reaches the AI provider.

Why not a browser extension?

A browser extension only protects the browser it's installed in, and a user can disable or remove it. Running as a system service means coverage isn't tied to one browser, and IT can deploy and manage it the same way they manage any other endpoint software.

Installing on Windows

Windows 10/11, 64-bit. Silent, unattended install via MSI:

msiexec /i SafePrompt.msi LICENSEFILE="license.json" VERIFYINGKEYFILE="verifying_key.hex" /qn

For fleet rollout across many machines, see Enterprise Deployment. For a single machine, see Installation or just start a free trial.