PipeLab builds open-source security tools for AI agents. The flagship product is Pipelock, an agent firewall that scans HTTP, WebSocket, and MCP traffic for secret exfiltration, prompt injection, and tool poisoning.

Origin

I run a plumbing company in Georgia. I built a homelab to manage the business with AI agents. When those agents had shell access, API keys, and unrestricted network, I went looking for something to stop them from leaking everything. Nothing existed. So I built Pipelock.

The homelab grew into a real lab: on-prem Kubernetes, GPU compute, and an open-source security tool that caught the attention of the OWASP community. PipeLab is the company behind it.

What We Ship

Pipelock

Open-source agent firewall. 8,800+ tests, 88% coverage, single binary. Apache 2.0.

Agent Egress Bench

143 adversarial test cases for validating any agent security proxy.

Pipelock Rules

Signed community detection patterns. Hot-reloadable YAML bundles.

OWASP Contributions

Coverage mappings for MCP Top 10, Agentic Top 10, MITRE ATLAS, EU AI Act.

Philosophy

Build it yourself. Understand it yourself. If it breaks, you know why. If it works, you know how.

No cloud dependencies. No VC funding. No growth hacking. Open-source core, honest pricing, real engineering.