A commercial multi-agent platform where developers publish agent definitions and users install instances — like an App Store for AI agents.
商业化多 Agent 平台。开发者发布 Agent 定义，用户安装实例 —— 如同 AI Agent 的 App Store。
MnEx is a platform for building, deploying, and monetizing AI agents. It originated from the OpenClaw personal multi-agent system and is evolving into an open platform where third-party developers supply the agent marketplace.
MnEx 是一个用于构建、部署和商业化 AI Agent 的平台。它源自 OpenClaw 个人多 Agent 系统， 正在演进为一个开放的 Agent 平台，Agent 市场由第三方开发者供给。
Publish agent and team definitions to the marketplace. Define capabilities, composition rules, and pricing. Receive revenue share when users install your agents.
在市场中发布 Agent 和 Team 定义。设定能力、编排规则与定价。 用户安装你的 Agent 后即可获得收益分成。
Browse and install agents into your Groups. Agents learn and adapt within your context. Manage permissions at the instance level.
浏览市场并将 Agent 安装到你的 Group 中。 Agent 在你的上下文中学习与适应。在实例级别管理权限。
Developers ship definitions; users install instances. This separation enables safe multi-tenancy — each Group runs its own instances with isolated memory.
开发者发布定义，用户安装实例。这种分离实现了安全的多租户架构—— 每个 Group 运行自己的实例，记忆完全隔离。
The platform is built on four orthogonal layers, each solving one problem. Layers depend top-down: Entity → Relation → Asset → Permission.
平台建立在四层正交原语之上，每层只解决一个问题。 从上往下依赖：Entity（实体）→ Relation（关系）→ Asset（资产）→ Permission（权限）。
Core chain: 核心链路： Person p_xxx ∈ Group g_xxx → bind Team t_xxx → contains Agent a_001, a_002
Four entity types with a unified six-prefix ID system, each 12 hex digits.
四类实体，统一六前缀 ID 体系，均为 12 位十六进制。
Three-layer identity stack: definition (published by developer), instance (created when Team is installed), and context (runtime, discarded after session).
三层身份栈：定义层（开发者发布）、实例层（Team 安装时创建）、 上下文层（运行时栈帧，session 结束即丢弃）。
The atomic entity. Self-registration activates a Person. Each Person automatically gets a default Group (single-person). Contact fields for reachability.
原子实体。自注册即激活。每个 Person 自动获得默认 Group（单人）。 包含联系方式字段。
A collection of Agents. Definition published by developer; instance created when a Group installs the Team. Team composition defines collaboration rules.
Agent 的集合。开发者发布定义，Group 安装时创建实例。 Team composition 定义协作规则。
A collection of Persons. Default single-person Group created automatically; user-created Groups for teams. Role-based permissions and member management.
Person 的集合。默认单人 Group 自动创建；用户可创建多人群组。 角色权限与成员管理。
| Prefix | Entity | 实体 | Layer | 层次 | Description | 说明 |
|---|---|---|---|---|---|---|
| ad_ | Agent Definition | Agent 定义 | Definition | 定义层 | Template in the marketplace | 市场中的模板 |
| a_ | Agent Instance | Agent 实例 | Instance | 实例层 | Runtime instance, within a Team | 运行时的实例，归属于 Team |
| td_ | Team Definition | Team 定义 | Definition | 定义层 | Package in the marketplace | 市场中的套餐 |
| t_ | Team Instance | Team 实例 | Instance | 实例层 | Runtime instance, bound to a Group | 运行时的实例，绑定到 Group |
| p_ | Person | Person | — | Natural person, self-registered | 自然人，自注册 | |
| g_ | Group | Group | — | Collection of Persons | Person 的集合 |
Foundational choices that shape the entire platform.
塑造整个平台基调的基础决策。
| Decision | 决策项 | Outcome | 结论 |
|---|---|---|---|
| User model | 用户模型 | Individual = single-person Group; unified data model | 个人 = 单人 Group，统一数据模型 |
| Market model | 市场模型 | App Store analogy — developers publish definitions, users install instances | App Store 类比——开发者发定义，用户装实例 |
| Entity model | 实体模型 | Agent / Person / Team / Group — four types | Agent / Person / Team / Group 四类 |
| ID system | ID 体系 | Six prefixes (ad_/a_/td_/t_/p_/g_), 12 hex digits | 六前缀（ad_/a_/td_/t_/p_/g_），12 位十六进制 |
| Architecture | 架构原则 | Four-layer primitives, top-down dependency | 四层原语，从上往下依赖 |
| Entity isolation | 实体隔离 | Group → Team → Agent, no cross-layer access | Group → Team → Agent，不跨层 |
| Memory isolation | 记忆隔离 | Four-link chain; cross-Group learning only via definition upgrades | 四层记忆链路；跨 Group 学习仅走定义升级 |
| Preference control | 偏好控制 | Field-level access tags, self-describing | 字段级 access 标签，自描述 |
| Permission default | 权限默认 | Open by default; Revoke tightens, Grant restores | 默认全开；Revoke 收紧，Grant 恢复 |
MnEx is in active development. The core architecture and specifications are being established, and the team is working toward the first implementation milestone.
MnEx 正在积极开发中，核心架构与设计规范持续迭代，团队正朝着首个实现里程碑推进。
The four-layer primitive architecture (Entity · Relation · Asset · Permission) and entity specifications are being refined. Design decisions are documented and iterated upon.
四层原语架构（Entity · Relation · Asset · Permission）与实体规范 持续完善中，设计决策已文档化并持续迭代。
We are actively recruiting engineers and designers across backend, AI/LLM, frontend, platform, DevOps, and security to build the founding team.
正在招募后端、AI/LLM、前端、平台、DevOps、安全等方向的工程师和设计师， 组建创始团队。
Building the core primitive engine, agent runtime, and developer toolchain. Committed to delivering a production-grade multi-agent platform.
推进核心原语引擎、Agent 运行时和开发者工具链的建设， 致力于交付生产级多 Agent 平台。
We are building the core team to turn the MnEx architecture into reality. Join us in shaping the future of multi-agent platforms.
我们正在组建核心团队，将 MnEx 架构变为现实。 加入我们，一起塑造多 Agent 平台的未来。
Design and build the four-layer primitive engine: Entity storage, Relation flows, Asset pipelines, and Permission enforcement. High-concurrency, multi-tenant data isolation.
设计与构建四层原语引擎：Entity 存储、Relation 流程、 Asset 管线、Permission 实施。高并发、多租户数据隔离。
Build the LLM abstraction layer: unified tool-calling across model providers, message format adaptation, context window management, prompt routing.
构建 LLM 抽象层：跨模型厂商的统一 tool-calling、消息格式适配、 上下文窗口管理、提示词路由。
Implement the agent execution engine: definition-to-instance lifecycle, three-layer identity stack, memory four-link chain, and workflow orchestration primitives.
实现 Agent 执行引擎：定义到实例的全生命周期、三层身份栈、 四层记忆链路、工作流编排原语。
Build the platform dashboard, agent marketplace UI, and developer portal. Craft the experience for browsing, installing, and managing agents across Groups.
构建平台控制台、Agent 市场界面与开发者门户。打造跨 Group 浏览、安装、管理 Agent 的用户体验。
Build the developer toolchain: agent SDK, CLI, local testing environment, definition schema validation, and publishing pipeline.
构建开发者工具链：Agent SDK、CLI、本地测试环境、定义规范校验、 发布流水线。
Infrastructure as code, multi-cloud deployment, observability stack, and cost optimization. Build the foundation for a reliable, scalable multi-agent platform.
基础设施即代码、多云部署、可观测性体系、成本优化。 为可靠、可扩展的多 Agent 平台打下基础。
Design and audit the permission system, data isolation boundaries, and agent sandboxing. Ensure cross-Group isolation and secure agent execution environments.
设计与审计权限系统、数据隔离边界与 Agent 沙箱。 确保跨 Group 隔离和安全的 Agent 执行环境。
Design the end-to-end experience: agent discovery, installation flow, Group management, and developer publishing tools. Shape how humans and agents collaborate.
设计端到端体验：Agent 发现、安装流程、Group 管理、开发者发布工具。 塑造人与 Agent 协作的方式。
Interested in the project or joining the team? We'd love to hear from you.
对项目感兴趣或想加入团队？欢迎与我们联系。
Roger Wang
Shanghai, Yangpu, East Bund · China中国 · 上海 · 杨浦 · 东外滩