Zero Node Privacy Firewall

Protect customer data before calling an AI model.

Redact sensitive values, send safe tokens to your model, and restore authorized values in the response without exposing the original data to the model provider.

Quick Start

Production requests use an API key. Ask for beta access, then call the redaction endpoint:

curl -X POST https://zero-node.com/api/v1/redact \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Email Jane at [email protected]",
    "options": {
      "replacement_mode": "token"
    }
  }'

The response contains safe text, findings, and a short-lived session for token-mode rehydration:

{
  "session_id": "0123456789abcdef0123456789abcdef",
  "redacted": "Email Jane at <EMAIL_1>",
  "findings": [
    {
      "type": "EMAIL",
      "token": "<EMAIL_1>",
      "confidence": 0.98
    }
  ],
  "expires_at": 1784822400
}

The public playground uses /api/v1/demo/redact. It requires no key, but it deliberately creates no reversible session.

AI-safe workflow

1. Redact

Send text or structured JSON to Zero Node before it crosses your trust boundary.

2. Call the model

Give the LLM tokenized content such as <EMAIL_1>, not the original value.

3. Rehydrate

Use the same API key and one-time session to restore authorized tokens in the model response.

curl -X POST https://zero-node.com/api/v1/rehydrate \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "session_id": "SESSION_FROM_REDACT",
    "text": "Draft ready for <EMAIL_1>"
  }'

Detection coverage

Deterministic detectors handle constrained formats. Private local models add contextual person, organization, and location detection. Deterministic spans take priority.

Contact and network

Email, phone, IP address, URL secrets, and supported API-key formats.

Identity and finance

SSN, cards, expiry, security codes, IBAN, account, customer, tax, passport, and medical IDs.

Contextual beta

Person, organization, and location. Evaluate contextual results against your own data.

Available replacement modes are token, mask, remove, hash, pseudonym, and partial.

First-release limits

Boundary	Current behavior
Request body	64 KB maximum for authenticated redact and rehydrate requests.
Public demo	2 KB text, rate-limited, and no reversible session.
Structured JSON	String values are scanned; keys and non-string values are preserved. Maximum depth is 64.
Contextual model	At most 16 string leaves per contextual request.
Token sessions	Short-lived, scoped to the creating API key, and consumed on first successful rehydrate.

Security architecture

Application code does not write request or response content to logs.
Contextual detection runs in a private local sidecar, not an external AI API.
Redis persistence is disabled for short-lived mappings and counters.
Token mappings are scoped to the API key that created them.
Zero Node reduces exposure; it does not guarantee detection of every sensitive value.

Read the Privacy Policy, Terms, or request beta access.