LoopX: A Lightweight State Kernel for Long-Running AI Agent Teams
A new open-source Python project called LoopX has been released, serving as an agent-agnostic state kernel for loop engineering. It helps manage durable goals, execution logs, and handoffs across different coding agents like Claude Code and Codex. As AI agents transition from single-turn tasks to complex, long-running autonomous workflows, managing state and handoffs becomes critical. LoopX addresses this by providing a passive, authoritative control plane to keep agent operations reviewable and restartable. LoopX is designed to be passive, meaning it does not execute code or run prompt chains itself, but instead manages state features like quota-aware auto-wake, executable to-dos, and evidence logs. It is agent-agnostic, allowing it to coordinate tasks across peer agent teams.
## BACKGROUND
Loop engineering is the practice of designing the feedback loop around an AI agent, detailing how it receives goals, takes action, and remembers state. Tools like Claude Code represent agentic coding tools that execute commands and edit codebases, which often require external state management when collaborating or running long tasks.