AI 开发工具 · 开源内核 AI development tool · Open Core

装上 skill，
让 AI 把需求变成应用 Install the skill,
let AI turn ideas into apps

GewuMind Boot（格物智）是一套 AI 开发工具：装上 gewumind-skill，你手上的 Claude Code / Cursor / opencode 就获得完整的后端开发能力，按「需求 → 设计 → 生成 → 发布」四阶段交付应用。它跑在一个 低代码后端容器 上——HTTP 接口用脚本编写、页面用组件文件维护、菜单/字典/配置全部在线管理。 GewuMind Boot is an AI development tool: install gewumind-skill and your Claude Code / Cursor / opencode gains full backend development capability, delivering apps through a 4-stage pipeline — requirements → design → generation → publish. It runs on a low-code backend container: HTTP APIs as scripts, pages kept as component files, menus/dictionaries/config managed online.

5 分钟上手Get started in 5 min 查看源码View source

v1.2.0 · 完整英文版 / 中英双语同包 / 语言 profile 切换 · 更新说明 v1.2.0 · complete English edition / bilingual package / language profiles · release notes

magic-apiSpring Boot 3MySQL MCPSa-TokenVue 3

gewumind-boot

# 1. 装上 gewumind-skill
$ ln -s /path/to/gewumind-skill ~/.claude/skills/gewumind-boot

# 2. 跑起后端容器
$ git clone https://github.com/gewumind/gewumind-boot
$ cd gewumind-boot
$ mysql -h127.0.0.1 -uroot -p gewumind-boot \
      < db/init/01-baseline.sql
$ export SPRING_DATASOURCE_PASSWORD='******'
$ mvn clean package -DskipTests
$ bash scripts/start.sh

✔ 后台     http://localhost:9089   （admin / 123456）
✔ Magic-Web http://localhost:9089/magic/web

# 3. 对 AI Agent 说：
"用 gewumind-boot 做一个设备管理模块，
  发布到 localhost:9089"

→ 需求 → 设计文档 → 生成代码 → 一键发布

# 1. Install gewumind-skill
$ ln -s /path/to/gewumind-skill ~/.claude/skills/gewumind-boot

# 2. Run the backend container
$ git clone https://github.com/gewumind/gewumind-boot
$ cd gewumind-boot
$ mysql -h127.0.0.1 -uroot -p gewumind-boot \
      < db/init/01-baseline.sql
$ export SPRING_DATASOURCE_PASSWORD='******'
$ mvn clean package -DskipTests
$ bash scripts/start.sh

✔ Admin     http://localhost:9089   (admin / 123456)
✔ Magic-Web http://localhost:9089/magic/web

# 3. Tell your AI Agent:
"Build a device management module with gewumind-boot,
  then publish it to localhost:9089"

→ Requirements → Design docs → Code → One-click publish

3步跑起后端steps to run

4阶段 AI 管线stage pipeline

83篇 AI 知识库AI knowledge docs

∞个 AI Agent 可接入AI agents supported

它是什么What it is

一个容器，装下整个后端One container for your whole backend

把接口、数据、权限、组件、工作流、消息、AI 能力装进一个开箱即用的 Spring Boot 应用，业务用「配置 + 脚本」表达，而不是重复写样板代码。APIs, data, auth, components, workflow, messaging and AI in one ready-to-run Spring Boot app — business logic expressed as config + scripts, not boilerplate.

接口即脚本APIs as scripts

基于 magic-api，在线编写 HTTP 接口，热更新、免编译、免重启，天然适配 AI 生成。Powered by magic-api: write HTTP endpoints online, hot-reload, no compile, no restart — ideal for AI generation.

页面即组件文件Pages are component files

页面以 .ms 组件文件存在于 data/{zh,en}/magic-api/component/，后端启动时读取并注册为全局组件。改完文件、刷新页面就生效，不需要重新打包。Pages live as .ms component files under data/{zh,en}/magic-api/component/; the backend reads and registers them as global components at startup. Edit a file, refresh — no repackaging.

在线管理一切Manage everything online

菜单、角色、用户、字典、模块、配置、图标、页面组件，全部后台可视化维护。Menus, roles, users, dictionaries, modules, config, icons and page components — all maintained in the admin UI.

能力清单Capabilities

开箱即用的企业级能力Enterprise features, out of the box

低代码后端容器Low-code backend

magic-api 脚本驱动接口，多数据源、事务、缓存、校验、分页一站式。magic-api scripted endpoints: multi-datasource, transactions, cache, validation, paging.

工作流引擎Workflow engine

内置请假/采购/合同/跨部门演示：条件分支、抄送、逐层驳回、按节点分组审批。Built-in demos: conditional branching, CC nodes, layered rejection, per-node approval dialogs.

报表插件Report plugin

拖拽式报表设计 + 导出 Excel，业务查询直接落地。Designer-based reports and Excel export, ready for business queries.

消息中台Messaging hub

Kafka / MQTT / TCP / WebSocket / SSE / Redis 统一接入，配置化开关。Kafka, MQTT, TCP, WebSocket, SSE and Redis — toggled by config.

安全与授权Security & license

登录/密码策略、Sa-Token 会话、SSO、离线 License（免费 + 商业双版本）。Login/password policy, Sa-Token, SSO, offline license (free + commercial editions).

监听调试Debug & listen

用真实页面的请求打断点，前后台联调所见即真实入参（magic-api 无此能力）。Break on real frontend requests — see actual parameters while integrating front and back (not in magic-api).

数据与缓存Data & cache

MySQL 多数据源、Redis 缓存与共享会话、Druid 连接池、MyBatis 兼容。MySQL multi-datasource, Redis cache & shared sessions, Druid pool, MyBatis compatible.

模块化菜单Modular menus

菜单/字典/图标/配置在线注册，模块化授权，权限精确到按钮。Register menus/dictionaries/icons/config online; module-based authorization down to buttons.

可集成的内核Embeddable kernel

可作为 Maven 依赖（com.gewumind:gewumind-boot）嵌入你自己的 Spring Boot 工程。Use as a Maven dependency (com.gewumind:gewumind-boot) inside your own Spring Boot app.

API 网关 · 开放接口API gateway · Open API

把后端开放出去，接 APP、小程序和第三方系统Expose your backend to apps, mini-programs and partners

内置一套完整的开放接口网关：注册应用 → 换 token → 按接口授权 → 限流 → 留调用日志。外部系统不碰你的后台账号体系，只拿一个应用身份调用被授权的接口。 A built-in open-API gateway: register an app → exchange a token → authorize per endpoint → rate-limit → keep call logs. External systems never touch your admin accounts; they call authorized endpoints with an app identity.

应用注册App registration

为每个接入方建一个应用，拿到 appKey 与 appSecret；secret 可随时重置，停用即刻失效。Create one app per partner with an appKey and appSecret; secrets can be rotated and apps disabled instantly.

token 鉴权Token auth

用 appKey + appSecret 换取 token，之后请求带 Authorization: Bearer <token>，由拦截器统一校验。Exchange appKey + appSecret for a token, then send Authorization: Bearer <token> on every call — validated by a shared interceptor.

按接口授权Per-endpoint grants

授权精确到「路径 + 方法」，还能给单个接口设独立的速率限制——小程序只给读接口，后台管理接口不开放。Grants are scoped to path + method, each with its own rate limit — give the mini-program read-only endpoints and keep admin APIs closed.

# 1. 用 appKey / appSecret 换 token
$ curl -X POST http://localhost:9089/open/app/token \
    -d 'appKey=xxx&appSecret=yyy'

# 2. 带 token 调用被授权的接口
$ curl http://localhost:9089/api/device/list \
    -H 'Authorization: Bearer <token>'

# 未授权 / 超限 / token 失效 → 各自返回对应状态码

# 1. Exchange appKey / appSecret for a token
$ curl -X POST http://localhost:9089/open/app/token \
    -d 'appKey=xxx&appSecret=yyy'

# 2. Call an authorized endpoint with the token
$ curl http://localhost:9089/api/device/list \
    -H 'Authorization: Bearer <token>'

# unauthorized / over limit / expired token → each returns its own status code

典型对接场景Typical integrations

APPMobile app 微信小程序WeChat mini-program 第三方系统Third-party systems 硬件网关Device gateways 合作方后台Partner portals 开放平台Open platform

每次调用都会记录来源 IP、请求参数、响应状态与耗时，可在后台按应用查询；开放接口的清单还能直接导出成文档交给对接方。 Every call is logged with source IP, request params, response status and cost time, queryable per app; the endpoint list can be exported as docs for partners.

AI 快速开发AI rapid development

一个 skill，把 AI 变成你的开发团队One skill turns any AI agent into your dev team

gewumind-skill 把 GewuMind 的 AI 开发能力打包成任何 Agent 都能用的 skill，内置完整知识库与强制模板，走可验收的四阶段管线。 gewumind-skill packages GewuMind's AI development capability into a skill any agent can use, with a full knowledge base, enforced templates and a verifiable 4-stage pipeline.

需求文档Requirements

生成 docs/requirement.md，确认业务边界。Produce docs/requirement.md and confirm scope.

设计文档Design docs

产出 ui / database / api，经校验门退出门。Write UI / database / API docs, gated by a verifier.

代码生成Code generation

产出 SQL / magic-script / Vue SFC / 菜单字典。Generate SQL, magic-script, Vue SFCs, menus & dictionaries.

一键发布One-click publish

通过 MCP bundle_apply 发布到后端。Publish to the backend via MCP bundle_apply.

适配任何 AgentWorks with any agent

支持 Claude Code、Cursor、opencode、deepseek-harness 及任意 MCP 客户端；无 MCP 时走 REST 契约。Supports Claude Code, Cursor, opencode, deepseek-harness and any MCP client; falls back to a REST contract without MCP.

Claude CodeCursoropencode deepseek-harnessMCP

安装（以 Claude Code 为例）Install (Claude Code example)

$ ln -s /path/to/gewumind-skill ~/.claude/skills/gewumind-boot

$ claude mcp add --transport sse gewumind-boot \
    http://localhost:9089/mcp

其他 MCP 客户端可把 SKILL.md 贴进项目 rules，stdio 用 npx mcp-remote 桥接。MCP 与发布接口都需要 X-AI-Key，开发模式启动时会自动生成并打印在日志里。For other clients, paste SKILL.md into project rules; bridge stdio with npx mcp-remote. Both MCP and the publish API need an X-AI-Key; in dev mode one is generated at startup and printed to the log.

查看 gewumind-skillExplore gewumind-skill

架构Architecture

AI 在前，内核在后AI up front, kernel behind

你的 AI AgentYour AI agent Claude Code · Cursor · opencode · deepseek-harness · 任意 MCP 客户端 Claude Code · Cursor · opencode · deepseek-harness · any MCP client

↓

gewumind-skill 四阶段管线 + 83 篇知识库 + 强制模板 4-stage pipeline + 83 knowledge docs + enforced templates

↓ MCP / REST

GewuMind Boot magic-api 容器 · 页面组件 · 工作流 · 报表 · 权限 · AI magic-api container · page components · workflow · report · auth · AI

↓

MySQL

Redis

Kafka / MQTT

TCP

WS / SSE

视频教程Video tutorials

跟着视频，5 分钟跑起来Watch it run in five minutes

一条总览 + AI coding + 三种部署，中文旁白 + 字幕。点右侧任意一集切换。One overview, AI coding in practice, and three deployment paths. English narration with subtitles — pick an episode.

在 B 站看总览Overview on Bilibili 在 YouTube 看英文版English overview on YouTube

EP 01

总览：你不想拥有一个编程大师吗？Overview: don't you want a coding master?

为什么说它不是低代码，而是 AI coding：Agent 直接写代码、直接发布。Why it's AI coding, not another low-code platform — the agent writes and ships the code.

快速开始Quick start

三步，跑起来Three steps and it runs

只想跑起来做业务开发？直接下 gewumind-boot-app 发行包解压即用，无需 Maven / Node；想改内核再走下面的源码流程。前置条件：JDK 17、Maven 3.8+、MySQL 8。Prerequisites: JDK 17, Maven 3.8+, MySQL 8.

# 1. 克隆 + 建库并导入基线（45 张表）
$ git clone https://github.com/gewumind/gewumind-boot.git
$ cd gewumind-boot
$ mysql -uroot -p -e "CREATE DATABASE gewumind-boot
      DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci"
$ mysql -uroot -p gewumind-boot < db/init/01-baseline.sql

# 2. 注入数据库密码（仓库里不写死）
$ export SPRING_DATASOURCE_PASSWORD='******'

# 3. 打包并启动（停止用 bash scripts/stop.sh）
$ mvn clean package -DskipTests
$ bash scripts/start.sh

# 1. Clone, create the database and import the baseline (45 tables)
$ git clone https://github.com/gewumind/gewumind-boot.git
$ cd gewumind-boot
$ mysql -uroot -p -e "CREATE DATABASE gewumind-boot
      DEFAULT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci"
$ mysql -uroot -p gewumind-boot < db/init/01-baseline.sql

# 2. Inject the database password (never hard-coded in the repo)
$ export SPRING_DATASOURCE_PASSWORD='******'

# 3. Package and start (stop with bash scripts/stop.sh)
$ mvn clean package -DskipTests
$ bash scripts/start.sh

后台Adminhttp://localhost:9089

Magic-Webhttp://localhost:9089/magic/web

默认账号Default loginadmin / 123456

MySQLlocalhost:3306 · gewumind-boot

改端口 SERVER_PORT=9090 bash scripts/start.sh；改数据库用环境变量 SPRING_DATASOURCE_URL / _USERNAME / _PASSWORD 覆盖。 Change port with SERVER_PORT=9090 bash scripts/start.sh; override the datasource via SPRING_DATASOURCE_URL / _USERNAME / _PASSWORD.

配置数据库Configure the database

基线脚本不代劳建库：先用 CREATE DATABASE 建出 gewumind-boot（字符集 utf8mb4），再把 db/init/01-baseline.sql 导入进去 —— 图形化客户端的「运行 SQL 文件」或命令行都行，导入后是 45 张表加平台自带的菜单/角色/字典/页面组件。 No bootstrap script creates the database for you: run CREATE DATABASE for gewumind-boot (charset utf8mb4) first, then import db/init/01-baseline.sql into it — a GUI client's "run SQL file" or the command line both work. You end up with 45 tables plus bundled menus, roles, dictionaries and page components.

# application-dev.yml（dev profile 默认值）
spring:
  datasource:
    url: jdbc:mysql://localhost:3306/gewumind-boot?...
    username: root
    password: ******

# 环境变量覆盖（推荐用于部署）
$ SPRING_DATASOURCE_URL=jdbc:mysql://db:3306/app \
  SPRING_DATASOURCE_USERNAME=app \
  SPRING_DATASOURCE_PASSWORD=****** \
  bash scripts/start.sh

# application-dev.yml (dev profile defaults)
spring:
  datasource:
    url: jdbc:mysql://localhost:3306/gewumind-boot?...
    username: root
    password: ******

# override with environment variables (recommended for deployment)
$ SPRING_DATASOURCE_URL=jdbc:mysql://db:3306/app \
  SPRING_DATASOURCE_USERNAME=app \
  SPRING_DATASOURCE_PASSWORD=****** \
  bash scripts/start.sh

支持哪些数据库：目前仅 MySQL 8。发行包里只内置了 mysql-connector-j 一个 JDBC 驱动，建库基线也是 MySQL 语法。 Supported databases: MySQL 8 only. The distribution ships a single JDBC driver, mysql-connector-j, and the baseline schema is MySQL syntax.

要接 PostgreSQL / Oracle / SQL Server，需要自行添加对应驱动依赖，并把 db/init/01-baseline.sql 转成目标方言（表选项、字符集、自增与注释语法均不通用）。 For PostgreSQL, Oracle or SQL Server, add the driver dependency yourself and port db/init/01-baseline.sql to that dialect (table options, charset, auto-increment and comment syntax are all MySQL-specific).

生态Ecosystem

四件套，各司其职Four pieces, one story

Core

GewuMind Boot

AI 驱动的低代码后端容器，项目的运行内核。The AI-powered low-code backend container — the runtime kernel.

Skill

gewumind-skill

装进任意 AI Agent 的开发技能：四阶段管线 + 知识库 + 强制模板。The dev skill you install into any AI agent: 4-stage pipeline + knowledge base + enforced templates.

Upstream

magic-api

内核底座的来源，开源低代码接口引擎。The open-source low-code API engine this is built on.

Community

社区Community

Star、Fork、提 Issue，一起把 GewuMind 做得更好。Star, fork and open issues — help make GewuMind better.

开源仓库Open source

gewumind-boot · 一个仓库，直接可用gewumind-boot · one repo, ready to use

这是承载内核的应用壳工程：内核以 jar 依赖的形式随仓库入库，配置、数据库、脚本工作区各归各位，clone 下来三条命令就能跑起一整套低代码后端。 This is the application shell that hosts the kernel: the kernel ships as a jar dependency inside the repo, with config, database and script workspace in their proper places — three commands from clone to a running low-code backend.

仓库里有什么What's in the repo

libs/ 内核 jar（thin jar + pom，标准 Maven 仓库布局，随工程入库）libs/ the kernel jar (thin jar + pom in a standard Maven layout, checked in)
data/{zh,en}/magic-api/ 中英文接口脚本与页面组件工作区，改完即生效data/{zh,en}/magic-api/ the Chinese/English workspace for API scripts and page components — edits take effect live
db/init/01-baseline.sql 建库基线：43 张系统表 + 平台自带菜单/角色/字典/组件数据，自己建库后导入即可db/init/01-baseline.sql the baseline: 43 system tables plus bundled menus, roles, dictionaries and page components — create the database yourself, then import this file
src/main/java/ 启动类与少量适配代码，业务逻辑不写在这里src/main/java/ the bootstrap class and a little glue — business logic does not live here
scripts/ 启停与证书脚本（Windows 同名 .bat 行为一致）scripts/ start/stop and certificate scripts, with matching Windows .bat equivalents

四种用法，随你挑Four ways to use it

下载发行包直接用：从 Releases 下载 gewumind-boot-app，解压后建库导入基线，bash scripts/start.sh 即跑，无需 Maven / NodeDownload & run: grab the gewumind-boot-app release, unzip, import the baseline, run bash scripts/start.sh — no Maven / Node required
直接跑：clone 后建库 + 打包 + 启动，得到一个完整的低代码后端Just run it: clone, init the database, package, start — a complete low-code backend
嵌进已有工程：把 com.gewumind:gewumind-boot 作为 Maven 依赖加进你的 Spring Boot 服务Embed it: add com.gewumind:gewumind-boot as a Maven dependency in your own Spring Boot service
交给 AI：装上 gewumind-skill，让 Agent 按四阶段把需求变成能跑的应用Let AI do it: install gewumind-skill and let an agent turn requirements into a running app

github.com/gewumind/gewumind-bootgithub.com/gewumind/gewumind-boot

版本与授权Editions & license

免费起步，按需升级Start free, upgrade when you need

免费版Free

开源 · 永久Open · forever

系统设置 / 登录 / 密码策略System settings / login / password policy
用户、角色、菜单、字典Users, roles, menus, dictionaries
主题、水印、自动重启、多数据源Themes, watermark, auto-restart, multi-datasource
基础 CRUD 与页面组件Basic CRUD & page components
开放接口网关（APP / 小程序 / 第三方对接）Open-API gateway (apps, mini-programs, partners)

商业版Commercial

离线 License 解锁Unlocked by offline license

SSO 单点登录SSO single sign-on
工作流引擎Workflow engine
模块管理Module management
Redis / Sa-Token 共享会话Redis / Sa-Token shared sessions
MQTT · Kafka · WebSocket · SSE · TCPMQTT · Kafka · WebSocket · SSE · TCP

按机器计费 · 随部署规模与支持需求定制Per-machine pricing · scoped by deployment & support

30 天免费试用，License 由签发中心离线签发、绑定机器指纹。30-day free trial; licenses are issued offline and bound to a machine fingerprint.

获取报价 / 申请试用Request a quote / trial gewumind@foxmail.comgewumind@foxmail.com

装上 skill，让 AI 把需求变成应用 Install the skill,let AI turn ideas into apps