Hi, I'm Mariano, building secure AI agents for enterprises in Philly

I'm Mariano Mattei, founder of Mattei Systems in Philadelphia. We help enterprises with AI transformation and cybersecurity: agentic AI engineering, cloud security, and compliance-ready workflows. I start with the business goals, data, and risk picture, then build agents for one job each, with clear evaluation criteria before anything reaches production.

My main open-source project is ClaudePaw, a self-hosted platform that turns a Claude subscription into a team of specialized agents. It includes a multi-project dashboard, seven security layers, semantic memory, a cron scheduler, and five messaging channels. I'm also building paw-trader, an autonomous trading system with risk gating, kill switches, and Telegram approvals. It's early, so I'm learning as much from my own testing as from how others run these tools.

I'd love feedback on the security model in ClaudePaw from anyone running multi-agent setups in production. Tell me where the seven layers would break for your use case.