AgentFence

AI coding agents have full filesystem and network access on your developers' machines. They can read .env files, access production credentials, and hit internal APIs with zero audit trail. Agent Safehouse just hit 685 points on HN because developers desperately want protection — but it's macOS-only with no team management.

Cross-platform agent sandboxing with a team dashboard. Define what each agent can access, get audit logs of everything it touched, and set org-wide policies. No more wondering if Claude Code read your AWS keys. SOC2-ready exports. Works with Claude Code, Cursor, Copilot, and any MCP-connected agent.