RedMatiq Privacy Redactor

I'm a solo developer in Switzerland building RedMatiq, a native Mac app that strips personal data from documents before you send them to AI.

The problem is straightforward: professionals paste contracts, medical records, and financial documents into ChatGPT every day. Most know they shouldn't. Few have a practical alternative to spending 30 minutes manually blacking things out.

RedMatiq sits in between. Drop in a PDF, Word doc, or spreadsheet. A BERT model running locally on your machine detects names, addresses, phone numbers, financial identifiers, and replaces them with consistent pseudonyms: "John Smith" becomes <PERSON_1>, "Zurich" becomes <LOCATION_1>. Same entity, same placeholder, across every document in a project.

You review what it found, toggle off false positives, then export a clean file or pipe it into an AI chat where the model never sees the original data. The response comes back with real names restored. To the user it reads naturally; to the AI, the original data never existed.

The stack: SwiftUI frontend, Python backend (Presidio + spaCy + a multilingual BERT NER model via HuggingFace). Document parsing with Docling. Everything runs locally, no cloud calls during processing. The only time data leaves your machine is when you explicitly send the pseudonymized version to an AI.

Why I'm building this

I work in an environment where people handle sensitive documents daily. The tension between "this AI could save me two hours" and "I can't risk this data" is constant. People either avoid AI entirely or quietly take the risk. Both outcomes are bad.

I wanted a tool that removed the dilemma. Not by trusting the AI vendor to handle data responsibly, but by making it structurally impossible for them to see the real data in the first place.

Where I am now

The app is in closed beta with a small group of testers. Core workflow works: import → detect → review → export or chat. Multi-document projects with consistent entity mapping across files. An MCP server that lets you use the same privacy layer with Claude Desktop or Cursor.

I'm looking for feedback on two things:

1. Is this a painkiller or a vitamin? I can build the best redaction tool in the world, but if people don't feel enough pain to change their workflow, it doesn't matter. The beta is partly a test of that.

2. What's the right price? Currently thinking about a one-time purchase (no subscription). The product is aimed at professionals - lawyers, consultants, compliance people - where the cost of a data leak dwarfs any tool price. But I'm aware that pricing a desktop app in 2026 is its own adventure.

What I've learned so far

• Local-only processing is a hard sell until it isn't. Most people don't think about where their data goes until the first time they paste something they shouldn't have. The "aha moment" for testers was seeing the raw text that would have been sent to the AI versus what RedMatiq actually sends.

• "Privacy tool" is a terrible category. People agree privacy matters. Almost nobody searches for privacy tools. I spent months positioning this as a redaction product before realizing the pull is on the other side: people don't want to protect data, they want to use AI freely. The category isn't privacy, it's AI enablement with a safety net. I'm still working through what that means for the product and the messaging, but the direction is clear.

• I still don't know if this is a painkiller or a vitamin. The professionals who manually redact documents before sharing them get it immediately. But that group is often not on Mac, and frequently can't buy software without IT approval. The much larger group pastes sensitive data into ChatGPT without thinking twice. They have the problem but don't feel the pain. The beta is largely a test of which group is the real market.

The business model

One-time license (not subscription). Price TBD based on beta feedback, somewhere between €59 and €149. There's also a pay-per-use relay service: a European-hosted AI model you can use without setting up API keys, billed by message. BYOK is planned for people who want to use their own keys.

No VC, no employees, no plans for either. Revenue = downloads × conversion. The math works at a few hundred licenses per year.

If you want to try it

I'm looking for beta testers, especially people who regularly handle sensitive documents and wish they could use AI on them without the risk. It's macOS only because running ML models locally needs real hardware resources, and supporting Windows packaging, signing, and distribution as a solo dev isn't realistic right now.

If that sounds like you, I'd genuinely appreciate the feedback!