ZeroEgress SDK
Why client-side changes your compliance story
Every existing PII-scrubbing solution, Presidio, cloud APIs, middleware, runs server-side. By the time it scrubs anything, the raw PII has already entered your infrastructure: your TLS terminators, your logs, your queues, your liability.
An SDK that runs in the end user's browser moves the boundary. The sensitive data is detected and redacted before the request leaves the user's device. Your backend only ever sees clean text. For your compliance team, that's not a smaller attack surface, it's a different category of answer.
What it would look like
const ze = await ZeroEgress.init(); // WASM + local model
const { clean, tokens } = await ze.redact(userInput);
// send `clean` to your API / LLM, PII never left the browser
const restored = ze.reveal(llmResponse, tokens);
What's in the box
- Detection engine: NER (local model) + 17 pattern categories + custom patterns, with second-pass propagation
- Reversible tokenization: consistent [NAME_1]-style tokens with a local-only reveal map, built for LLM round-trips
- Audit chain: HMAC-SHA256 tamper-evident event log with ECDSA-signed JSON/CSV exports and a signed compliance certificate, give your compliance team a verifiable record
- Deployment modes: full (local AI model) or lightweight (patterns only, no model download)
- Licensed per application, not per seat. Covered by US Patent 12,665,758 and additional applications pending.
Built for
- AI chat and copilot interfaces that pass user text to LLM APIs
- Support and CRM tools where agents paste customer conversations
- Healthcare and legal frontends that must keep PHI/PII out of the backend
- Browser extensions and internal tools handling sensitive text
Early access
The SDK is in development. We're selecting a small group of design partners to shape the API, early partners get direct access to the team, influence over the roadmap, and preferential license terms.
Interested? Write to with the subject "SDK early access" and one line about what you're building. That's the whole form, we don't do signup databases, for reasons that should be obvious by now.