
Brakeproof
Safety layer for AI agents — action firewall, spend brake, proof of work & red-team tests
My role: Founder — designed and built end to end
- Action Firewall
- Red-Team Tests
- Mobile Approvals (beta)
Brakeproof is my own product: a safety layer between AI agents and the systems they work on. Every tool call passes a checkpoint, so safe actions run, risky ones wait for human approval with a snapshot taken first, and dangerous ones are blocked. It caps LLM spend, checks “done” claims against Stripe, GitHub or the database, keeps a tamper-evident ledger and red-team tests LLM apps and agents against OWASP, NIST AI RMF and MITRE ATLAS. Built on FastAPI and Next.js with Python and TypeScript SDKs, an MCP proxy and a React Native approvals app (beta); works with Claude Code, MCP, OpenAI Agents SDK and LangGraph. Deploy to your own cloud is in beta. In private preview with a waitlist.
- FastAPI
- Python
- Next.js
- React
- TypeScript
- React Native
- MCP
- AI
- LLM Security




















