为中国坐标系而生的
Web 地图基础设施 Mapping infrastructure for the modern web, built for China's coordinate systems

一套地图 API，跑在 OpenLayers 或 MapLibre GL 任一引擎上。

One mapping API that runs on either OpenLayers or MapLibre GL.

• 统一门面 createMap({ engine })，可按子路径锁定引擎、摇掉另一个；引擎独有能力走 getEngine() 逃生舱
• 高德、百度等国内底图就地纠偏（GCJ-02 / BD-09），对外坐标恒为 WGS-84
• 框架无关的编辑引擎：25 个编辑命令、撤销重做与宏撤销、开洞 / 切面 / 合并等拓扑工具、吸附辅助线、三路合并同步
• Vue 与 React 官方绑定；8 个包，依赖方向由门禁强制

• Unified facade createMap({ engine }); lock an engine by subpath import and tree-shake the other; engine-only features through the getEngine() escape hatch
• Chinese basemaps (GCJ-02 / BD-09) corrected in place — your coordinates always stay WGS-84
• Framework-agnostic editing engine: 25 commands, undo/redo and macro undo, holes / split / merge topology tools, snap guides, three-way-merge sync
• Official Vue and React bindings; 8 packages with dependency direction enforced by lint gates

基于 GeoVerse SDK 的协作式地理数据编辑平台。

A collaborative geodata editing platform built on GeoVerse SDK.

• 渲染引擎是项目的一个属性：OpenLayers ↔ MapLibre 一键切换，图层、样式与协作数据完全通用
• 20+ 绘制与拓扑工具、逐版本历史与回滚，乐观锁冲突检测 + 字段级三路合并
• 专题制图（自然断点 / 分位数 / 归一化 / 色觉友好色带）、实时协作与要素评论、AI 助手
• 免登录分享与 iframe 嵌入（可加密传输）；多租户 RBAC；Docker Compose 私有化部署

• The rendering engine is a project property — switch OpenLayers ↔ MapLibre; layers, styles and collaboration data carry over
• 20+ drawing and topology tools, per-version history and rollback, optimistic-lock conflict detection with field-level three-way merge
• Thematic mapping (natural breaks, quantiles, normalisation, CVD-safe ramps), real-time co-editing with feature-anchored comments, an AI assistant
• Login-free sharing and iframe embeds (with encrypted transport); multi-tenant RBAC; private deployment with Docker Compose

轻量地理数据分发服务——单二进制、纯 Go、零外部运行时依赖。

A lightweight geodata server — a single, pure-Go binary with zero external runtime dependencies.

• PostGIS（ST_AsMVT 下推）、MySQL 8 / MariaDB、PMTiles、MBTiles、GeoJSON、GeoPackage → MVT、WMTS、TileJSON、OGC API – Features
• 内置数据源控制台：真实连接测试、配置原子写回与热加载，数据库凭据不回显
• 两级缓存、ETag / 304、gzip、可选 API Key；-doctor 只读部署诊断
• MCP 端点让 LLM 直接发现图层、查询要素；算法插件：最短路径、等时圈、路径匹配、DBSCAN

• PostGIS (ST_AsMVT push-down), MySQL 8 / MariaDB, PMTiles, MBTiles, GeoJSON, GeoPackage → MVT, WMTS, TileJSON, OGC API – Features
• Built-in data-source console: real connection tests, atomic config write-back with hot reload, credentials never echoed
• Two-tier cache, ETag / 304, gzip, optional API keys; -doctor read-only deployment checks
• An MCP endpoint lets LLMs discover layers and query features; algorithm plugins: shortest path, isochrones, map matching, DBSCAN

AI-native 空间应用运行时：人点 UI、AI 调工具、Agent 自主执行、外部 MCP 客户端，走同一个内核。

An AI-native runtime for spatial apps: UI clicks, AI tool calls, autonomous agents and external MCP clients share one kernel.

• 能力描述符就是工具定义：同一份投影同时驱动 UI 命令面板、AI 工具目录与 MCP tools/list
• 一切操作走单一漏斗 dispatcher.invoke：权限、校验、审计、事务日志回放长在内核里
• 领域状态撤销是一等公民：工作流合并成一次宏撤销；写操作可 dryRun 预览，交给人审批
• 内核只依赖 zod、不认识"地图"；自然语言不进内核——425 个测试无需真实 LLM

• A capability descriptor is a tool definition — one projection feeds the UI command palette, the AI tool catalogue and MCP tools/list
• Every operation goes through one funnel, dispatcher.invoke: permissions, validation, audit and journal replay live in the kernel
• Domain-state undo is first class: workflows collapse into one macro undo; writes can dryRun for human approval
• The kernel depends only on zod and doesn't know what a map is; natural language never enters it — 425 tests run without a real LLM

Spring Boot starter，打通 MyBatis Plus 与 JTS 几何类型。

A Spring Boot starter bridging MyBatis Plus and JTS geometry types.

• MySQL 8 与 PostgreSQL / PostGIS 自动识别，零配置
• Point 到 GeometryCollection 共 8 类几何注解，WKB 自动转换
• Jackson GeoJSON 序列化开箱即用，REST 接口直接吐 GeoJSON
• Java 17+，Spring Boot 2.7+ / 3.x

• Auto-detects MySQL 8 and PostgreSQL / PostGIS — zero configuration
• Annotations for 8 geometry types from Point to GeometryCollection, WKB conversion built in
• Jackson GeoJSON serialisation out of the box for REST APIs
• Java 17+, Spring Boot 2.7+ / 3.x

面向 AI Agent 的空间技能仓库，配套确定性布局引擎与 MCP App。

Spatial skills for AI agents, with a deterministic layout engine and an MCP App.

• generate-region-layout：把设施网格 / 矩形与摆放、编号规则变成确定性、可校验的布局——货架、库位、车位、展位
• 配套 npm 包 @geoverse/layout-engine 0.7 与 @geoverse/layout-mcp-app 0.5：交互地图卡片 + 离线 HTML 预览
• 每个 Skill 带评测用例与安装自检；统一 CI 做结构校验与干净安装冒烟

• generate-region-layout: turn facility grids or rectangles plus placement and numbering rules into deterministic, validated layouts — racks, bins, parking bays, booths
• Backed by @geoverse/layout-engine 0.7 and @geoverse/layout-mcp-app 0.5 on npm: interactive map cards plus an offline HTML preview
• Every skill ships evals and installation self-checks; one CI validates structure and runs clean-install smoke tests