Show HN：Legit，基于 Git 的开源 AI 智能体版本控制
3 分•作者: jannesblobel•6 个月前
大家好，我们是 Martin、Nils 和 Jannes。<p>我们正在构建 Legit，一个面向 AI 智能体和 AI 原生应用的开源版本控制和协作层。<p>可以在这里找到代码库 <a href="https://github.com/Legit-Control/monorepo" rel="nofollow">https://github.com/Legit-Control/monorepo</a>，以及网站 <a href="https://legitcontrol.com" rel="nofollow">https://legitcontrol.com</a><p>在过去的几年里，我们开发了多个开发者工具和 AI 驱动的产品。一旦我们开始让智能体修改真实文件和关键业务数据，一个问题就不断出现。我们无法可靠地回答发生了什么变化、为什么发生变化，或者如何安全地撤销它。<p>如今，大多数 AI 工具要么在没有真正保护措施的情况下运行，要么将其状态存储在难以检查、审计或迁移的专有数据库中。一旦智能体开始在共享数据上进行协作，你通常只能祈祷一切顺利。<p>我们注意到了一些有趣的事情。开发人员在协作编写代码时没有这个问题，而类似智能体的工作流程首先在那里兴起。原因相对简单。Git 已经解决了协调、历史记录、审查和回滚的问题。<p>这一见解促使我们构建了 Legit。我们将 Git 风格的版本控制和协作引入到 AI 应用程序和大多数文件格式中。智能体所做的每一次更改都会被跟踪。每一个操作都是可检查、可审查和可逆的。没有隐藏状态。没有黑盒历史记录。<p>Legit 作为轻量级 SDK 工作，AI 应用程序可以在文件系统运行的任何地方嵌入它。它处理智能体的版本控制、同步、回滚和访问控制。所有内容都存储在一个存储库中，你可以自己托管，也可以托管在任何你已经信任的 Git 托管提供商上。<p>我们认为扩展 AI 协作的正确方法不是隐藏智能体的行为，而是让开发人员和用户能够查看、审查和控制每一次更改。Legit 是我们尝试将现代开发者工作流程的纪律性、可见性和安全性带入到支持写入的 AI 应用程序中。<p>试用一下：<a href="https://github.com/Legit-Control/monorepo" rel="nofollow">https://github.com/Legit-Control/monorepo</a>，并让我们知道你的反馈、批评和想法。
查看原文
Hi HN, Martin, Nils, and Jannes here.<p>We are building Legit, an open source version control and collaboration layer for AI agents and AI native applications.<p>You can find the repo here <a href="https://github.com/Legit-Control/monorepo" rel="nofollow">https://github.com/Legit-Control/monorepo</a> and the website here <a href="https://legitcontrol.com" rel="nofollow">https://legitcontrol.com</a><p>Over the last years, we worked on multiple developer tools and AI driven products. As soon as we started letting agents modify real files and business critical data, one problem kept showing up. We could not reliably answer what changed, why it changed, or how to safely undo it.<p>Today, most AI tools either run without real guardrails or store their state in proprietary databases that are hard to inspect, audit, or migrate. Once agents start collaborating on shared data, you are often just crossing your fingers and hoping nothing goes wrong.<p>We noticed something interesting. Developers do not have this problem when collaborating on code, and agent like workflows took off there first. The reason is relatively simple. Git already solves coordination, history, review, and rollback.<p>That insight led us to build Legit. We bring Git style versioning and collaboration to AI applications and to most file formats. Every change an agent makes is tracked. Every action is inspectable, reviewable, and reversible. No hidden state. No black box history.<p>Legit works as a lightweight SDK that AI apps can embed anywhere the filesystem works. It handles versioning, Sync, rollback, and access control for agens. Everything lives in a repository that you can host yourself or on any Git hosting provider you already trust.<p>We believe the right way to scale AI collaboration is not to hide what agents do, but to let developers and users see, review, and control every change. Legit is our attempt to bring the discipline, visibility, and safety of modern developer workflows to write enabled AI applications.<p>Give it a spin: <a href="https://github.com/Legit-Control/monorepo" rel="nofollow">https://github.com/Legit-Control/monorepo</a> and let us know your feedback, criticism, and thoughts.