Turnstone brings cluster-wide AI agents to DevOps

Most AI agent frameworks ignore the hard part — running agents safely across production infrastructure at scale — and Turnstone fills that gap with a security-first design that treats agent autonomy as a governance problem.

• –Ships 15 built-in tools (bash, file ops, web fetch, memory, planning, task spawning) plus MCP external tool servers, giving agents everything needed for real sysadmin work
• –Two-tier "LLM judge" validates every tool call before approval with a risk score — a rare explicit safety gate in the agent-for-ops space
• –Cluster mode load-balances agents across nodes via Redis, with directed per-node routing for targeted operations; ZeroClaw and OpenClaw lack this tier
• –BSL-1.1 license converts to Apache 2.0 in 2030 — free to self-host, restricted only for managed-service resale
• –Low current traction (2 upvotes, 0 comments on Reddit), but the architecture suggests a serious project aimed at platform and DevOps engineers, not hobbyists