Where your data goes

The first question about any tool that handles personal data is whether its developer can see the data. Here the answer is not a promise but the way the app is built, and you can check it.

Why the developer cannot receive your data

The app has no code that sends data over the network. On start it also blocks network connections for its own process, and an automated test in the open source code checks this.

The language models are part of the app and run on your Mac's processor. They need no cloud service.

The mapping between placeholders and real data is kept only on your Mac, in an encrypted vault (AES-256). The key is in the macOS Keychain of this Mac, so a copy of the vault cannot be opened on another computer. The protection depends on the protection of your Mac account.

Check it yourself in a minute

1. Turn off Wi-Fi or unplug the cable.
2. Open pii-roundtrip and drop in a document.
3. The app works as usual. It does not need the internet.

For a stricter check, install a network monitor such as the free LuLu or Little Snitch. It shows that pii-roundtrip opens no network connections.

Open source

All the code of the app and of the benchmark is published on GitHub under the Apache 2.0 licence. Your security team can read every line and build the app themselves. Vulnerabilities can be reported privately under the security policy.

What does reach the AI tool

The AI tool receives the cleaned text. The app replaces direct identifiers in it: names, NHS and National Insurance numbers, bank details, addresses. It can miss some: in our benchmark at least one value remained in 0.2% of the test documents, so read the text before you send it. Indirect details such as a diagnosis, a job title or unusual circumstances stay in the text. If they could identify the person, remove them by hand before you send it.