huangruiteng/loopx

Traditional agent frameworks focus on single-session tasks and rely on chat memory or timers, which are insufficient for long-running, multi-agent, multi-day objectives. LoopX solves the problem of maintaining stable control state (objectives, gates, todos, evidence, quota) across turns, tools, and agents, ensuring work remains reviewable, restartable, and handoff-friendly without requiring a central leader or autonomous production control.

Developers can build systems for multi-day engineering, research, benchmark, or experiment objectives; issue and PR loops that preserve scope and evidence; recurring heartbeat or monitor workflows; and peer-agent teams with ownership, leases, and handoff. The ceiling includes orchestrating complex, long-horizon tasks with human-in-the-loop gates, quota-based scheduling, and verifiable evidence trails, as demonstrated by 200+ hour public contribution arcs. It is not designed for fully autonomous production control, but rather for bounded, reviewable agent slices.