Permanent redaction for compliance. Reversible sanitization for AI workflows. One API, two modes, zero exposure.
# Redact — permanent PII removal: clean = ambientmeta.sanitize("Patient: John Smith, SSN 123-45-6789", mode="redact") # clean.text → "Patient: [PERSON_1], SSN [SSN_1]" (PII gone forever) # Sanitize — reversible for LLM workflows: safe = ambientmeta.sanitize("Summarize John Smith's file at john@acme.com") reply = your_llm(safe.text) # LLM sees placeholders, never real data final = ambientmeta.rehydrate(reply, safe.session_id) # restores originals
Permanent redaction for compliance. Reversible sanitization for AI. Same API, different outcomes.
PII is permanently removed. No session created. No recovery possible. The data is gone. Ship it to auditors, store it in logs, share it freely.
PII is replaced with safe placeholders. Send the clean text to any LLM. Restore original values when you need them back. Sessions expire after 24 hours.
Two API calls. That's it.
Send us text, we replace PII with safe placeholders.
"Email John at john@acme.com"
→ "Email [PERSON_1] at [EMAIL_1]"Send the safe text to Claude, GPT-4, Gemini—whatever you want.
reply = your_llm(safe.text)Send us the LLM response, we restore original values.
"Sure, I'll email [PERSON_1] at [EMAIL_1]"
→ "Sure, I'll email John at john@acme.com"Detect and strip PII directly in ChatGPT, Claude, and Gemini. No API key needed. No code changes. Install, sign in with Google, and go.
Standard PII, healthcare identifiers, and anything custom you define.
Define any regex pattern via the /v1/patterns endpoint
Every correction teaches the system something new about your data.
PII identified via regex, NER, and compiled rules
You flag what we got wrong
Contradictions reveal ambiguity patterns
Rules are validated and deployed
Runtime accuracy increases automatically
When you correct a misclassification — a phone number that's actually an NPI — the system detects the contradiction, extracts what distinguishes the two contexts, formulates a disambiguation rule, validates it against known-good examples, and compiles it into the runtime engine. Next time, it gets it right. No retraining. No model updates. Deterministic rules compiled from your corrections.
First-class Python SDK. Drop-in wrappers for popular frameworks.
pip install ambientmetapip install langchain-ambientmetapip install llama-index-ambientmetaDrop-in wrapperDrop-in wrapperfrom langchain_ambientmeta import PrivacyGateway gateway = PrivacyGateway(api_key="am_live_xxx") safe_llm = gateway.wrap(your_llm) # Use normally — PII handled automatically response = safe_llm.invoke("Summarize the employee file")
Start free. Pay as you grow. No credit card required.
Enterprise-grade redaction for compliance, legal, and government teams. Page-based pricing starting at $199/month. Same detection engine, built for document workflows.
Data sovereignty without building infrastructure. Or deploy on your own — same API, air-gapped, data never leaves your network.
Sensitive data stays in your control. The LLM only sees placeholders.
Self-hosted option for maximum control. Deploy in any region.
Detailed logs for every request. SOC 2 Type II in progress.
All of them. Claude, GPT-4, Gemini, Llama, Mistral—we're model-agnostic. We work at the text level, so any LLM that accepts text input works.
95%+ for standard entities out of the box. Our intelligence system learns from your corrections — flag a mistake, and the system formulates disambiguation rules so it doesn't happen again. Custom patterns can achieve 99%+.
Designed for compliance. PII never touches external APIs. We detect healthcare identifiers (NPI, MRN, DEA) out of the box. Self-hosted option for maximum control. SOC 2 Type II certification in progress.
Less than 20ms p50 for sanitize, less than 5ms for rehydrate. Fast enough that users won't notice.
Yes. Use the /v1/patterns endpoint to add detection for employee IDs, project codes, or any org-specific identifiers via regex.
Yes. Same API, your infrastructure. Single Docker image. Air-gapped deployments supported. Contact sales for pricing.
Create your account in 30 seconds. No credit card required.
Get Started Free