Know what your AI tools actually touch and send

The problem

You installed the assistant, the agent in your editor, a handful of MCP servers, because they are useful. Each one can read files and reach the network, and none of them shows you what it actually did.

An agent with a filesystem tool and a fetch tool can read anything and send it anywhere, and the first you would hear of a problem is after it happened. You would not run a background process with that much access and no logging. That is more or less what everyone is doing with AI tools right now.

What Provae for AI does

• Watches every AI tool call on your machine. It sits between your AI client and its tools and records what ran, which files were touched, and whether anything sensitive was involved. Locally. Nothing leaves.
• Flags tampering. If a tool you approved quietly changes what it does, you get an alert. This is the most common way an AI agent turns against you.
• Keeps a record you own. A local, exportable log, signed so you can confirm it has not been altered: what ran, which categories of sensitive data were touched, and whether anything was tampered with. Yours, on your machine, for whatever you need it for.

Zero collection, and you can check

Provae for AI stores everything in a local file you own. It has no server to send your activity to. The sensitive details it inspects, it reads in memory and never writes down; the record holds categories and counts, never your file paths or their contents. You do not have to trust that. The tool is built so you can confirm it.