LoopX

LoopX is a local-first state kernel and control plane for long-running AI agent work. The official source for this OpenTools record is https://github.com/huangruiteng/loopx. The listing is written from the public repository and project docs, not from a directory snippet, so the claims stay close to the README, package metadata, and visible project positioning. The working model is straightforward: builders keep objectives, gates, executable todos, evidence logs, quota, claims, leases, and handoffs in durable state while Codex, Claude Code, Cursor, OpenCode, or shell agents execute bounded slices of work. That matters for builders because the hard part is rarely a one-turn demo. Teams need a repeatable install path, clear state boundaries, and enough operational detail to decide whether the tool belongs in a local workflow, a team workspace, or a controlled infrastructure environment. Key capabilities include durable goals, scope tracking, human gates, executable todos, evidence logs, quota-aware auto-wake, peer-agent claims, leases, typed continuation, run history, and verifiable handoffs. These features are useful when a team wants to give agents a safer place to work, keep long-running work visible, or make environment setup less dependent on a single laptop. The project is especially relevant for developers experimenting with coding agents, hosted workspaces, and human-reviewed automation. Best fit: developers and small teams running multi-hour or multi-day agent work where chat history, reminders, and ad hoc notes are not enough to keep progress reviewable. A solo developer can use it to test a workflow without waiting on procurement. A platform team can evaluate it as an open-source component. A larger organization should still run access-control review, security review, data handling review, and model-cost review before connecting it to private repositories or production systems. Pricing from the public source is simple: the repository is MIT licensed and public; users still pay for their own model providers, machines, schedulers, or connected infrastructure. That does not mean every deployment is cost-free. Users can still pay for cloud compute, model APIs, storage, hosted runners, GPUs, databases, or third-party services connected to the workflow. Start with a small test and check the official docs before relying on a specific provider or command. Why it stands out: it focuses on the control-state layer rather than trying to be another model provider or all-in-one agent framework, which makes it useful alongside existing coding agents. It has a clear AI-builder use case, public implementation details, and enough project surface area to evaluate from source. Treat it as an engineering component: verify installation, run one low-risk workflow, inspect the outputs, then expand only after the access boundaries and operating costs are predictable.