
Improvements in the security model
We’re currently refining Rectury’s security model around least-privilege access to production infrastructure. The current approach combines dedicated non-admin users, scoped filesystem access, policy-gated execution, approval boundaries for privileged operations, and restricted host-level capabilities, while keeping broader SSH access explicitly opt-in. The goal is to make the agent operationally useful without giving it unnecessary authority over the rest of the server. I’m curious how people running real production infrastructure view this tradeoff, and whether this is the level of isolation and control you’d expect before allowing an AI agent to operate on a VPS.