⚡ New: Protect your data from both sides of AI with our unified perimeter defense.

Blindfold PII Proxy

Sanitize prompts before they reach external AI models. We automatically detect, tokenize, and restore sensitive data so third-party LLMs never see your users' real identities.

Start Free

How Tokenization Works

AIGuard sits between your application and your LLM provider, ensuring PII is stripped outbound and accurately restored inbound.

// 1. The original prompt leaves your server
"My name is John Doe and my SSN is 000-11-2222. Am I eligible?" ↓ Intercepted by AIGuard Blindfold Proxy // 2. Blindfold detects PII, generates tokens, and forwards to the LLM
"My name is [NAME_001] and my SSN is [SSN_001]. Am I eligible?" ↓ LLM generates response using tokens // 3. LLM replies to AIGuard
"Yes, [NAME_001]. We have verified [SSN_001]." ↓ AIGuard securely unmasks tokens in volatile memory // 4. Final response delivered to your user
"Yes, John Doe. We have verified 000-11-2222."

Deterministic Token Mapping

We replace your data with strict semantic tokens (like [EMAIL_001]) and force the LLM to preserve them via injected system prompts. Context is maintained without sacrificing privacy.

Memory-Only Processing

Your raw prompt data and PII tokens are processed entirely in volatile memory. We never write your unmasked data to a database or disk, ensuring total compliance.

Seamless SSE Streaming

Blindfold handles split tokens over Server-Sent Events natively. You can stream responses directly to your users chunk-by-chunk without leaking partial PII.

Zero-Latency Routing

Our Python proxy processes regex pattern matching and asynchronous logging via Celery workers, adding virtually zero overhead to your existing LLM latency.

Full & Partial Masking Modes

Choose between full anonymization for strict environments, or unlock Partial Masking (Pro Tier) to selectively hide specific data types based on your exact requirements.

Denial of Wallet Protection

Hard-coded quota checks at the edge evaluate your tier limits before requests are forwarded. If limits are hit, we drop the request, protecting your provider billing.

One Line of Code. Total Security.

You don't need to rewrite your pipelines or learn a new SDK. Just point your existing LLM client to AIGuard's secure proxy.

# Update your environment variables to point to the AIGuard Proxy
export LLM_BASE_URL="https://api.aiguard.solutions/v1"
# Keep your existing LLM provider keys configured as normal
Start Building Free