Scan every prompt — and every model response — for emails, phones, SSNs, IBANs, contracts and custom entities, before and after they reach OpenAI, Anthropic or Gemini. One REST call or one MCP tool, P95 under 80ms in both directions.
Not another regex library. A hybrid engine that combines deterministic rules with NLP models, tuned for enterprise LLM traffic — applied to the prompt that goes in and to the response that comes back.
Edge-deployed detection engine, applied to both directions — prompt in and response out. No noticeable latency added to your LLM calls.
Rule-based patterns for structured entities plus NLP models for names, addresses and unstructured PII.
Get back tokens instead of masked text. Reveal originals later via controlled RPC with full audit.
Standard REST endpoint or Model Context Protocol tool for agents. Same detection engine either way.
Send text, get structured entities with type, position, confidence and an aggregate risk score.
POST https://api.privaro.ai/v1/detect
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json
{
"text": "Contact John Doe at john@acme.com or +34 600 123 456"
}
// Response
{
"entities": [
{ "type": "PERSON", "value": "John Doe", "start": 8, "end": 16, "confidence": 0.97 },
{ "type": "EMAIL", "value": "john@acme.com", "start": 20, "end": 33, "confidence": 1.0 },
{ "type": "PHONE", "value": "+34 600 123 456", "start": 37, "end": 52, "confidence": 0.99 }
],
"risk_score": 0.72
}Pattern matching catches the easy entities. Contextual PII — names, addresses, medical conditions, contract clauses — requires language understanding. Privaro combines both in a single hybrid engine.
| Entity type | Regex | Privaro hybrid NLP |
|---|---|---|
| Email addresses | ||
| Credit cards / IBANs | ||
| Full names (contextual) | ✗ | |
| Addresses | ✗ | |
| Medical conditions | ✗ | |
| Contract clauses | ✗ | |
| Spanish DNI / NIE | Partial | (with checksum) |
| Custom entities | ✗ | (configurable) |
Every detection result includes the entity type, exact position, and a confidence score — ready to route through your pipeline's policy engine.
Use /v1/detect to inspect a prompt, or /v1/proxy/protect to tokenize sensitive entities and forward the sanitized text to the LLM in one call.
// Option A: detect only (inspect what's in a prompt)
POST /v1/detect
{ "text": "Call Maria at maria@company.com" }
// Option B: protect + relay (tokenize and forward to the LLM)
POST /v1/proxy/protect
{
"text": "Call Maria at maria@company.com",
"pipeline_id": "YOUR_PIPELINE_ID",
"provider": "openai",
"model": "gpt-4o"
}
// Response includes the LLM answer with tokens re-identified
{
"result": "I'll contact Maria at maria@company.com",
"entities_detected": 2,
"risk_score": 0.68,
"audit_log_id": "log_abc123"
}Pairs natively with the Privaro AI Governance Platform for policy enforcement, token vault and audit trails.
Free tier includes 10k detections/month. No credit card required.
Request API access