Data Protection
Your data never reaches the AI.
Nema uses automatic pseudonymization to replace sensitive entities before they leave your infrastructure. The AI works with placeholders — your real data stays with you.
Why pseudonymization?
Traditional AI tools process your data as-is. Nema takes a different approach: sensitive information is detected and replaced with consistent tokens before anything reaches the model.
- Names, companies, and personal data are replaced before any API call to the AI model.
- The AI delivers the same quality — it reasons about structure and context, not specific identifiers.
- After the response, pseudonyms are mapped back to originals in your browser. The round-trip is invisible.
- Custom entities let you protect internal terms that automated detection might miss.
How it works
Three steps, zero data exposure
Every message passes through a pseudonymization pipeline before it reaches the AI. The mapping is maintained per session, so the model can reason coherently across a conversation without ever seeing real data.
Detect
Your message is scanned for sensitive entities: names, companies, products, locations, and your custom terms.
Replace
Each entity gets a consistent pseudonym (e.g. PER_x7k2). The same entity always maps to the same token within a session.
Restore
After the AI responds, pseudonyms are mapped back to originals in your browser. Only you see the real data.
Example
Schedule a call with Sarah Miller from Acme Corp about the Atlas project.
···
The mapping lives in your browser. The AI model only processes the pseudonymized version.
Detected entity types
In short
Pseudonymization is not an opt-in afterthought — it is a core layer of how Nema processes data. When enabled, every message is cleaned before it leaves your infrastructure, and every response is restored transparently. The AI never sees the real thing.