Privacy infrastructure for the AI era
Your AI doesn't need
to know your name.
Every time you paste a document into an AI, your personal data — names, addresses, bank details, health records — leaves your device forever.
PrivacyLayer strips all personal information before it reaches any AI model. Everything runs in your browser. Nothing is sent to our servers. Nothing.
Chrome Extension now available
Works directly inside ChatGPT, Claude and Gemini — no copy-paste needed.
50+
Data types detected
AES-256
Vault encryption
0 bytes
Sent to any server
6
Compliance frameworks
The problem
Without PrivacyLayer
You paste a hospital invoice into ChatGPT. Your name, your diagnosis, your insurance number, your home address — all of it travels to a server in the US. Stored. Logged. Gone.
AHV: 756.1234.5678.97
IBAN: CH93 0076 2011 6238 5295 7
→ sent to AI provider servers
With PrivacyLayer
The same invoice. But the AI only sees placeholders. It still gives you a useful answer -- and we swap the real values back in afterward. The AI never knew your name.
AHV: [AHV_01]
IBAN: [IBAN_01]
→ AI responds → real values restored
How can you be sure your data stays private?
All scanning, redaction, and encryption happens in JavaScript running in your browser tab. Open your browser's Network tab — you will see zero outgoing requests with personal data. There is no backend that processes your documents. The source code is fully open on GitHub. This is not a promise — it is a technical guarantee you can verify yourself.
Why this matters now
Too many AI models
Claude, GPT, Gemini, Mistral, Llama — new models launch every week. You want to test them all. But every model means another company seeing your data. PrivacyLayer lets you try any model safely.
Data sovereignty matters
European companies and citizens need to use AI — but can we trust US-based providers with our health records, financial data, and personal information? PrivacyLayer means you don't have to trust anyone.
Layers of privacy risk
Some data is obvious — names and SSNs. But what about "my brother lives in Zurich" or "I take insulin daily"? We detect both explicit identifiers and contextual personal information that can identify someone indirectly.
Data flow
The AI model (greyed) receives only anonymized text — it operates inside the same pipeline as any other vendor.
How it works
Upload
Drop in a PDF, snap a photo of a letter, or paste text. Your document enters the privacy layer right here in your browser — it never touches any server.
Detect
50+ detection patterns scan your document in parallel. Names, emails, Swiss AHV numbers, IBANs, phone numbers, addresses, dates of birth, insurance IDs — if it can identify someone, we catch it.
Replace
Every personal detail is swapped with a safe placeholder — [NAME_01], [EMAIL_01], etc. The mapping is locked in an AES-256 encrypted vault that only you hold the key to.
Share safely
The cleaned document is safe to send to any AI model, any colleague, any service. Nobody sees real names, real numbers, real addresses. Zero personal data leaves your device.
Restore
When the AI responds using placeholders, we swap the real values back in. You read a natural response with real names and real details. The AI never knew them.
Audit trail
Every scan and redaction is logged locally — what was found, what category, what risk level. Ready for GDPR, HIPAA, and SOC 2 compliance reviews.
What we detect and protect
Compliance
Use any AI without violating HIPAA.
HIPAA's Safe Harbor method defines 18 specific identifiers that must be removed for health data to be legally de-identified. PrivacyLayer detects and strips all 18 — names, dates, phone numbers, emails, account numbers, IP addresses, and more — before your text reaches any AI model. What the AI sees is not Protected Health Information. You stay compliant. The AI stays useful.
HIPAA Safe Harbor
All 18 HIPAA identifiers detected and replaced. De-identified output is not PHI and can be sent to any AI provider without violating HIPAA.
GDPR Article 4
No personal data leaves your device. GDPR only governs personal data — anonymized data is out of scope. PrivacyLayer anonymizes before transmission.
Swiss DSG / nDSG
Built with Swiss data standards in mind. AHV numbers, CH-IBANs, and Swiss address formats are detected by default.
Also relevant for teams working under CCPA, SOC 2, ISO 27001, and PCI-DSS requirements.
For developers
If you build AI-powered apps, PrivacyLayer drops into your existing code in one line.
import { wrapLanguageModel } from "ai";import { privacyLayer } from "privacylayer";// Before — personal data goes straight to the modelstreamText({ model, messages });// After — one line. Zero personal data leaves your app.streamText({model: wrapLanguageModel(model, privacyLayer()),messages,});