pipelock

Based on the available evidence, pipelock looks like a noteworthy early-stage open-source security project with a clear positioning, not a broadly validated enterprise platform yet. Adoption should be judged cautiously: GitHub growth from roughly 151 to 752 stars and repeated X reposts show attention, but that is heat proof, not usability proof. The stronger signals come from the official repo and the creator’s own explanations, which consistently frame it as an agent egress firewall that sits between agents and the internet or tools to inspect, block, and log actions.

Its practical role is not a generic “LLM firewall” or a cloud WAF. A better analogy is a security proxy/gateway layer for agentic tools like Claude Code, Cursor, or Codex. The evidence says it scans mediated HTTP, MCP, A2A, and WebSocket traffic, aiming to catch exfiltration, SSRF, prompt injection, and tool poisoning, while also offering signed action receipts, a kill switch, and what the author describes as a scanner-plus-sandbox design. In other words, it is focused on runtime egress control and auditability, not on model training governance or a basic API router.