Agentctl – Local AI Agent Control Plane

A lightweight Go tool that sits between AI coding agents and risky system actions (package installs, shell execution, file writes, API calls), letting developers define and enforce security policies locally without hosting infrastructure. Solves the problem of safely running autonomous agents on your machine while maintaining full control and auditability.

The surge in agentic coding tools — Cursor, Devin, Claude Code, and a dozen others — has made local sandboxing a genuine unsolved problem; developers are already duct-taping together Docker restrictions and shell wrappers because nothing purpose-built exists yet. No clear incumbent owns this space, though larger players like Docker and cloud security vendors could absorb it into existing products if the category gets traction. The $2k–10k/mo revenue band is credible only if it prices as a developer subscription or team license, since solo devs will balk at paying for something they assume the agent vendor should provide for free. The biggest risk is that the major agent platforms — Anthropic, Cursor, GitHub — simply ship native sandboxing, collapsing the wedge entirely before any meaningful adoption builds.