Z Privacy

Z Privacy

You know this. Every question to AI has a hidden cost: the information you send with it.

Use AI without moving sensitive information outside your device.

Before anything is sent, you are shown exactly what will leave your device — and nothing leaves without your word.

Linux and Windows builds, in beta

The idea

The problem we are trying to solve is: how can individuals and companies benefit from AI and external services without having to move sensitive information outside their devices or organizations? For example, reviewing and verifying tax-related documents can take a significant amount of time, while AI can perform much of this work within minutes. The challenge is how to use AI without exposing clients’ names, personal information, or financial data.

Z Privacy works as an information vault inside the user’s computer. During the workflow, sensitive information is kept locally on the device and replaced with identifiers that Z Privacy can recognize and restore later. This allows the relevant content to be sent to an AI service without sending the original sensitive information. Before anything is sent, the user is shown exactly what will leave the device, and nothing leaves without their word. The interaction can take place either through a chat interface or through documents such as PDF files. When the AI response is received, Z Privacy restores the original information locally on the user’s device. Everything Z learns is kept in an encrypted vault on the user’s own machine — even the vault file on disk contains no readable name. Z Privacy is not tied to one company: it speaks to OpenAI and to Anthropic, it lists xAI, DeepSeek, Moonshot and Google beside them, for the day a key for one of those four arrives, and every connection is made at an address the user can change — so any service that answers an OpenAI-compatible request can be used, including a model running on the user’s own computer, which needs no key at all.

The path

ImportScanReviewProtectChoose who answersAIRestore

As it appears in the app.

Z Privacy minimizes the sensitive content you send. It does not make you anonymous to the AI provider.

What an in-app chat sends, said plainly, because it is what leaves this machine: your question is protected, and that protected question is the whole of what travels. Asked again inside Z Privacy it travels with one thing added — the exchange before it, the previous question and its answer in the protected form they left in — and never the document, which is sent once, with the question that first carried it. Nothing older is sent, and nothing older is lost: every earlier turn stays in your session’s own sealed file. What «Copy the chat» puts on your clipboard is that same protected question and nothing earlier, because the chat you keep with a model of your own keeps its own thread; an answer you bring back by hand still restores every name the chat has ever carried. What this costs you is said here too: a follow-up reaches a model that can no longer see the document, and no door here hands anybody the whole conversation.

Privacy

Source and licence

Open source — Apache License 2.0

Use it as you like, in a company or alone, and ask nobody: the one condition is the one the licence already carries. A product built on this code says so where it lists what it is built on — «Z Privacy, Copyright 2026 Faruk AB», as the NOTICE file asks and Apache 2.0 §4(d) binds. That means a product’s own notices, its documentation or its «About» screen; it does not mean advertising, which is said here because it was asked and answered.