LoopX is a local-first control plane and research workbench for long-running AI agent work. The open-source project has about 6,400 stars on GitHub and was actively developed in 2026.
Long-horizon agent work has a memory problem: goals, decisions, and evidence get scattered across sessions and tools, and every new session starts half-blind. LoopX keeps that state in one place. It preserves goals, decisions, ownership, and evidence across sessions and across runtimes โ including Codex, Claude Code, and DeepSeek Harness โ and organizes work around an agent-native Kanban model, where agents claim tasks, pass gates, monitor progress, and write results back. A Personal Agent Workspace gives individuals a durable home for ongoing agent projects, a dashboard PWA keeps humans in the loop, and protected operations stay behind explicit human authorization rather than agent discretion. Trajectory analysis is a core part of the workbench: instead of only seeing a final answer, reviewers can inspect how an agent got there and where a long run went off track. That makes LoopX as much a research instrument for understanding agent behavior as a project-management layer for getting work finished.
LoopX is free and open source under the Apache-2.0 license. There is no paid cloud tier; the workbench runs locally.
Developers and small teams running long-horizon agent workflows who need continuity, evidence, and oversight across many sessions instead of one-off chats.