# rungent > rungent is an open-source framework that makes a product agent-native from the HTTP API it already has. The product annotates its OpenAPI 3.1 operations with `x-rungent`; from that one catalog rungent serves an in-app agent (AG-UI, with the product's own forms and approvals), an MCP server, a CLI, skill.md, llms.txt, ⌘K search and answers that cite the product's docs. Every call runs as the signed-in user, with the user's own credential and a signed assertion the product verifies. It ships as one npm package (`rungent`, with the `rungent` command), the image `ghcr.io/rungent/rungent-server` for any stack, and a Python kit (`rungent`). To integrate rungent, a coding agent installs the skill at https://rungent.acahti.dev/skills/rungent/SKILL.md and follows it. This site is itself a rungent host: https://rungent.acahti.dev/mcp, https://rungent.acahti.dev/skill.md, and the Playground at https://rungent.acahti.dev/playground/. ## Docs - [Overview](https://rungent.acahti.dev/overview.md): rungent is an open-source framework that makes a product agent-native from the HTTP API it already has. One annotated catalog gives you the in-app agent, MCP, a CLI, a skill, agent-readable docs, global search and cited answers, every one acting as the signed-in user. - [Quickstart](https://rungent.acahti.dev/quickstart.md): Hand the integration to your coding agent with one sentence, or do it by hand in a JavaScript API. Either way the result is the same five pieces. - [How it works](https://rungent.acahti.dev/how-it-works.md): The parts of rungent, where the runtime runs, what happens during one request, the protocol the UI speaks, and the words these docs use. - [The catalog](https://rungent.acahti.dev/catalog.md): Your OpenAPI document, with x-rungent on the operations agents may use, is the catalog. What an operation is, what its effect implies, how names become ids, how two calls chain, and how to check it all before you ship. - [Acting as the user](https://rungent.acahti.dev/acting-as-the-user.md): rungent has no account in your product. It calls your API with the signed-in user's own credential and a short-lived signed assertion that names the operation; your API checks both and authorizes the call as it would a click in your UI. - [Forms and approvals](https://rungent.acahti.dev/forms-and-approvals.md): Every change takes one path. A request to create or change something opens your page's own form, prefilled; the submission becomes an approval card bound to its exact arguments; only an approval runs the call. - [Search and knowledge](https://rungent.acahti.dev/search.md): One search runs across your own search endpoints and your docs. People use it in ⌘K, the agent searches with it before it answers, and every answer cites the hits it used. - [Surfaces](https://rungent.acahti.dev/surfaces.md): Where your users and their agents reach your product through rungent. The dock, ⌘K, MCP, your branded CLI, skill.md and llms.txt all draw from one catalog, act as the user, and ask before every write. - [JavaScript](https://rungent.acahti.dev/javascript.md): Embed rungent in a JavaScript API and put the dock in your web app. What createRungent takes, how to mount it on Hono, Express and Cloudflare Workers, which store to use, and the few dock props that matter. - [Python and other stacks](https://rungent.acahti.dev/python.md): Add rungent to an API in Python, Go, Java or any other language. Your API serves an annotated OpenAPI document, answers who is signed in and verifies a signed header; rungent serve runs beside it, and one reverse proxy puts both on your origin. - [Deploy](https://rungent.acahti.dev/deploy.md): Run rungent in production. The Docker image and every environment variable, the database, secrets, the reverse proxy, replicas, and the model chain rungent falls through when one model fails. - [Testing](https://rungent.acahti.dev/testing.md): Run rungent's test suite against your own product. Contract cases check every surface without a model; scenarios you write measure how well the agent does real tasks and whether it stays safe. The latest results of the Playground and reference integrations are at the end. - [x-rungent](https://rungent.acahti.dev/x-rungent.md): Every key of the x-rungent OpenAPI extension, its type, its default and what it does, with the rules a catalog must pass. - [HTTP API](https://rungent.acahti.dev/http-api.md): Every endpoint rungent serves on your origin, how each is authenticated, the AG-UI events a run streams, the interrupts and how to answer them, and the shapes of answers and errors. - [CLI reference](https://rungent.acahti.dev/cli-reference.md): Every command and flag of rungent, the command in rungent. Where it finds the host and token, the authoring commands, rungent serve, the files it keeps and its exit codes. - [Playground](https://rungent.acahti.dev/playground.md): A sandboxed cloud console that this site hosts and this site's own rungent drives. Create, stop and delete servers from the dock, the CLI or an MCP client. ## 中文文档 - [概览](https://rungent.acahti.dev/zh/overview.md): rungent 是一个开源框架,让产品从已有的 HTTP API 出发变成 Agent 原生。一份标注好的目录,就能给出产品内 Agent、MCP、CLI、skill、给 Agent 读的文档、全局搜索和带引用的回答,而且都以当前登录用户的身份工作。 - [快速开始](https://rungent.acahti.dev/zh/quickstart.md): 一句话把接入交给你的编码 Agent,或者在 JavaScript API 里自己动手。两种方式的结果一样,都是同样的五个部分。 - [工作原理](https://rungent.acahti.dev/zh/how-it-works.md): rungent 由哪些部分组成、运行时跑在哪里、一次请求经过哪些步骤、界面说的是什么协议,以及这些文档使用的术语。 - [目录](https://rungent.acahti.dev/zh/catalog.md): 你的 OpenAPI 文档,在允许 Agent 使用的操作上加上 x-rungent,就是目录。本页讲什么是操作、效果意味着什么、名称怎样变成 ID、两次调用怎样串起来,以及上线前怎样把这些全部检查一遍。 - [以用户身份调用](https://rungent.acahti.dev/zh/acting-as-the-user.md): rungent 在你的产品里没有账号。它带着当前登录用户自己的凭据,以及一份写明操作的短时签名断言调用你的 API;你的 API 校验两者,像处理界面上的一次点击一样授权这次调用。 - [表单与确认](https://rungent.acahti.dev/zh/forms-and-approvals.md): 每次变更只走一条路。创建或修改的请求会打开你页面自己的表单并预先填好;提交的内容变成一张绑定确切参数的确认卡;只有用户确认后才会执行调用。 - [搜索与知识](https://rungent.acahti.dev/zh/search.md): 一次搜索覆盖你自己的搜索接口和你的文档。用户在 ⌘K 里用它,Agent 在作答前用它搜索,每个回答都会引用它用到的结果。 - [入口](https://rungent.acahti.dev/zh/surfaces.md): 你的用户和他们的 Agent 通过 rungent 进入你产品的地方。面板、⌘K、MCP、带你品牌的 CLI、skill.md 和 llms.txt 都来自同一份目录,都以用户身份操作,每次写操作前都会先问。 - [JavaScript](https://rungent.acahti.dev/zh/javascript.md): 把 rungent 嵌入 JavaScript API,再把面板放进你的 Web 应用。本页讲 createRungent 的参数、如何挂载到 Hono、Express 和 Cloudflare Workers、该用哪种存储,以及面板中真正要关心的几个 prop。 - [Python 及其他技术栈](https://rungent.acahti.dev/zh/python.md): 给 Python、Go、Java 或任何其他语言写的 API 加上 rungent。你的 API 提供一份标注好的 OpenAPI 文档,回答当前登录的是谁,并校验一个签名请求头;rungent serve 在旁边运行,一个反向代理把两者放到同一个 origin 上。 - [部署](https://rungent.acahti.dev/zh/deploy.md): 在生产环境运行 rungent。本页讲 Docker 镜像和全部环境变量、数据库、密钥、反向代理、多副本,以及模型链:一个模型失败时 rungent 依次换下一个。 - [测试](https://rungent.acahti.dev/zh/testing.md): 用 rungent 的测试套件测你自己的产品。契约用例不调用模型,检查每个入口;你写的场景衡量 Agent 完成真实任务的效果,以及它是否安全。页末是演练场和参考集成的最新结果。 - [x-rungent](https://rungent.acahti.dev/zh/x-rungent.md): x-rungent OpenAPI 扩展的每个键:类型、默认值和作用,以及目录必须通过的校验规则。 - [HTTP API](https://rungent.acahti.dev/zh/http-api.md): rungent 在你的源站上提供的每个端点:各自如何认证,一次运行推送的 AG-UI 事件,有哪些中断、如何回答,以及回答和错误的结构。 - [CLI 参考](https://rungent.acahti.dev/zh/cli-reference.md): rungent 的每个命令和参数,即 rungent 中的命令。它从哪里找宿主和令牌、编写类命令、rungent serve、它保存的文件和退出码。 - [演练场](https://rungent.acahti.dev/zh/playground.md): 一个由本站托管、由本站自己的 rungent 驱动的沙箱云控制台。在面板、CLI 或 MCP 客户端里创建、停止和删除服务器。 ## Optional - [rungent skill for coding agents](https://rungent.acahti.dev/skills/rungent/SKILL.md) - [Changelog](https://rungent.acahti.dev/changelog.md): 0.5.1 is the latest release - [Full text of every doc](https://rungent.acahti.dev/llms-full.txt) - [Source](https://github.com/rungent/rungent)