Salus masks personal data on your device before it reaches ChatGPT, Claude, or Gemini, then restores it in the reply.
Before anything is sent, Salus shows you what it found. Mask it, or whitelist what's safe.
Scans documents and images before they upload. Names, faces, and IDs are removed first.
The AI works on placeholders. Salus restores the real values in the reply, on your machine.
Claude, ChatGPT and Gemini, on both their desktop apps and in the browser.
Start with the desktop app, or bring the same on-device masking to chat, your own code, images, and meetings.
Masks AI traffic locally before it reaches ChatGPT, Claude, or Gemini, then restores it in the reply. Runs on macOS, Windows & Linux.
Book a demoPrivate AI chat in your browser. Ask anything; sensitive data is masked before it ever leaves your session.
Try Salus ChatDrop the same PII masking into your own apps and workflows with a simple, fast API.
Book a demoRedact personal data inside images and documents before they reach an AI model.
Try Salus toolsPrivate meeting transcription and AI notes, with personal data kept out of the prompt.
Try Salus toolsSalus runs on your device, between your AI app and the provider. Personal data is masked before it's sent, then restored in the reply.
Contracts, reports, emails, with real names and details. No restrictions.
A small model on your device finds the sensitive data in milliseconds.
Claude, ChatGPT, or Gemini receives only [PERSON_1], never the real values.
Placeholders are mapped back on your machine. You see the full answer.
If your job runs on confidential information, Salus lets you use AI on it without that information leaving your hands.
Draft and review with client names, matters, and privileged detail kept on your machine.
Work through client financials, tax IDs, and account numbers without exposing them.
Summarize notes and records while patient details stay on the device.
Analyze statements and customer data without it reaching the model.
Screen CVs and handle employee data with names and salaries masked.
Use client documents and data without leaking who they belong to.
| Use the best AI | Data stays on your device | Real answers back | Low effort | |
|---|---|---|---|---|
| Ban AI | people work around it | |||
| Redact by hand | easy to miss things | slow, manual | ||
| Law-specific LLM | weaker models | |||
| Cloud redaction tool | data hits their cloud | |||
| Salus |
Detection is local, only your AI traffic is touched, and Salus is a data-minimization measure, not a compliance product.
On-device detection. A small model runs on your device. Your prompts aren't sent to Salus or any server.
Only AI traffic. Banking, email, and everything else pass through untouched.
Keys and logs stay local. The certificate is generated on your machine, and nothing is written to disk in plaintext.
Data minimization (Art. 32). Masking identifiers before they reach the AI is a recognized measure that reduces what's exposed.
Not another processor. Because data never leaves your device, Salus doesn't become a new processor of your client data.
What it doesn't do. It doesn't replace your DPA with the AI provider, and it isn't a compliance certification.
Salus reduces how much sensitive data reaches AI providers and is designed to support your data-protection efforts. It doesn't make AI use automatically compliant.
Runs on your device. Free to start, and your sensitive data never leaves your device.
Rolling it out to a team? Deploy across everyone with a simple record of what's masked, metadata only. Talk to us →