Four layers of protection

AI Security That Works
at the Proxy Level

Every request and response passes through four security layers — redaction, policy enforcement, circuit breakers, and AI audit — before anything reaches the LLM or your users.

Secure Your AI TrafficRead the Docs

How the security pipeline works

Your App
sends request
Redaction
masks PII & secrets
Policies
blocks threats
Circuit Breaker
enforces limits
LLM Provider
processes request
Audit
reviews response
Your App
gets safe response

Four pillars of AI security

Each layer addresses a different threat vector — together they provide comprehensive protection.

Pillar 1

Redaction Engine

PII never reaches the LLM

Automatically detects and masks emails, SSNs, credit card numbers, phone numbers, IP addresses, and 100+ API key formats before the request leaves your network.

  • Regex-based PII detection with configurable categories
  • Secret detection for AWS, Stripe, GitHub, and more
  • Response redaction scans LLM output before it reaches your users
  • Custom patterns for domain-specific sensitive data
  • Allowlisting for values that should pass through
Pillar 2

Policy Firewall

Block dangerous commands at the gate

A rule engine that inspects every request against configurable blocklists. Dangerous shell commands, sensitive filesystem paths, and custom patterns are blocked before they reach the model.

  • Default blocklists for destructive commands (rm -rf, DROP TABLE, etc.)
  • Sensitive path protection (/etc/passwd, .env, SSH keys)
  • Custom regex rules with allow/block/flag actions
  • Per-key and organization-level policies
  • Full audit log of every blocked request
Pillar 3

Budget Kill-Switch

Runaway sessions can't run away

Real-time session tracking with configurable circuit breakers. Set limits on cost, tokens, duration, and request count. Loop detection catches repeated prompts and kills stuck agents.

  • Per-session cost, token, duration, and request limits
  • Loop detection for duplicate requests and cost spikes
  • Automatic session termination (auto-kill) on breach
  • Live session dashboard with manual kill button
  • Configurable thresholds per virtual key
Pillar 4

The Auditor

AI audits AI — every response reviewed

A second LLM reviews every response for exposed secrets, command injection, data exfiltration, and prompt injection. Git snapshots capture your codebase state at request time for full traceability.

  • Git snapshot webhooks capture code state per request
  • Double-check AI with configurable audit model
  • Async mode for zero-latency or blocking mode for maximum safety
  • Risk scoring: low / medium / high / critical
  • Cost-based and pattern-based audit triggers

Security features by plan

Every plan includes basic protection. Upgrade for advanced features.

FeatureFree
$0
Pro
$29/mo
Enterprise
$99/mo
PII redaction
Secret detection
Response redaction
Custom patterns
Policy firewall5 rulesUnlimited
Session tracking
Loop detection
Auto-kill
Git snapshots
Double-check AI
Blocking audit
One-click presetsRelaxedStandardStrict
Get Started Free