Cloakr

An enterprise AI gateway that sits between a company's data and the models it uses: it redacts sensitive data before anything leaves, routes each request to the cheapest capable model across 14 providers, and keeps a tamper-evident audit trail for ISO 27001, SOC 2 and GDPR reviews. I built the whole product — gateway, SDK and dashboard, every one of its 55 commits — toward a YC application and pilot customers, while my teammate ran business development.

Highlights

• Real-time PII detection with risk-graded highlighting and confidence scores, redacting before any prompt leaves the network
• Policy engine enforcing per-team budgets, data-residency rules (US/EU) and content filtering on every request
• Envelope encryption through AWS KMS with key rotation that never breaks old reads, and signed audit logs that detect tampering
• One API across 14 model providers — OpenAI, Anthropic, Azure, Bedrock, Mistral, Groq and more — with automatic routing and a live cost-savings view
• Built for production: JWT and OAuth sign-in, IP allow-listing, CSRF protection, quotas, rate limiting, Prometheus metrics and k6 load tests