huangruiteng/loopx

What it solves

LoopX provides a stateful control plane for long-horizon AI agents, solving the problem of maintaining consistency, governance, and progress across long-running tasks that span multiple days, sessions, or different agent harnesses. It prevents long-term goals from being lost in chat memory and ensures that work remains reviewable and restartable.

How it works

LoopX acts as a lightweight state kernel that runs on top of existing agent runtimes (like Claude Code, Cursor, or Codex) rather than replacing them. It maintains a durable record of objectives, gates, todos, and evidence. The process follows a cycle: LoopX determines if an agent should run based on quota and gates; the agent executes a bounded turn; and the LoopX state is updated with evidence and the next todo, making the state authoritative rather than the agent's session memory.

Who it’s for

It is designed for developers and researchers conducting multi-day engineering, research, or benchmark objectives, as well as teams using peer-agent systems where ownership, handoffs, and safety gates are required for long-running workflows.

Highlights

• Provider-Neutral: Works across various agent harnesses (Codex, Claude Code, Cursor, etc.) without being locked into one.
• Local-First Workspace: A personal dashboard for managing goals, tasks, and evidence that persists across restarts.
• Governance and Recovery: Implements gates for human judgment and quota systems to prevent uncontrolled agent spending.
• Long-Horizon State: Tracks evidence and decision lineage over hundreds of elapsed hours of project time.