用一个协议
定义 企业级 AI 能力
One Protocol to Define Enterprise AI
MxJSON 是开源的 AI 能力描述协议。将 API 接口、提示词、工作流、UI 渲染统一封装为一个 JSON 文件，让企业级 AI 能力像容器镜像一样可被定义、封装与分发。 MxJSON is an open protocol that unifies API definitions, prompts, workflows and UI rendering into a single JSON file. Package and distribute enterprise AI capabilities like container images.
{
"$mxjson": "1.0",
"$id": "contract.review",
"meta": {
"name": "合同智能审核",
"author": "@awei",
"category": "legal"
},
"input": {
"document": { "type": "file", "accept": ".pdf,.docx" }
},
"ai": {
"provider": "claude",
"model": "claude-opus-4",
"prompt": "分析合同风险条款...",
"temperature": 0.2
},
"workflow": [
"extract", "analyze", "review:human"
],
"output": {
"risks": "array",
"score": "number",
"summary": "string"
}
}{
"$mxjson": "1.0",
"$id": "medical.triage",
"meta": {
"name": "AI 智能分诊",
"category": "healthcare"
},
"input": {
"symptoms": { "type": "text", "required": true },
"vitals": { "type": "object" }
},
"ai": {
"provider": "qwen",
"model": "qwen-max",
"prompt": "基于症状推荐科室..."
},
"output": {
"department": "string",
"urgency": "enum",
"recommendations": "array"
}
}由一位坚定的工程师，
为下一个十年的 AI 而设计Designed by an engineer,
for the next decade of AI
罗孔威 Luo Kongwei · Awei Luo Kongwei 罗孔威 · Awei
"AI 能力不应被锁在某个平台或某个 SDK 里。
它应该像 JSON、像 Docker 镜像一样，可被任何系统读懂、运行、分发。" "AI capabilities shouldn't be locked behind a single platform or SDK.
They should be readable, runnable and distributable—like JSON, like a container image."
罗孔威（Awei），上海派许网络科技有限公司/MianxiangAI创始人，MxJSON 协议发起人与核心规范作者。长期从事企业级 AI 平台架构，主持设计了 MxJSON 协议从规范、运行时引擎、人机协作审核到资产分发的完整体系。他相信：协议先于产品，标准先于规模。 Luo Kongwei (Awei) is the co-founder of Shanghai Zhiwen Chuangyue AI Technology Co., Ltd., and the founder & lead author of the MxJSON protocol. With years of experience building enterprise AI platforms, he led the design of MxJSON's specification, runtime engine, human-in-the-loop review system and asset distribution layer. His thesis: protocol before product, standard before scale.
六个维度，统一标准化Six Dimensions of Standardization
一个 JSON 协议，覆盖从 API 生成到 UI 渲染的全流程。One JSON protocol covering everything from API generation to UI rendering.
API 自动生成API Generation
从协议定义自动生成 RESTful 接口、OpenAPI 文档与部署配置，零样板代码。Auto-generate RESTful endpoints, OpenAPI docs and deployment configs from a single spec. Zero boilerplate.
提示词模板Prompt Templates
支持变量注入、版本管理、A/B 测试的标准化提示词，告别散落的字符串。Standardized prompts with variable injection, versioning and A/B testing. No more strings scattered in code.
流程编排引擎Workflow Engine
串行 / 并行 / 分支 / 循环，原生支持人机协作（Human-in-the-loop）审核节点。Serial, parallel, branching and looping logic with first-class human-in-the-loop checkpoints.
一键分发One-click Distribution
像容器镜像一样跨平台部署。公有云、私有化、边缘端，一份资产到处运行。Deploy across platforms like a container image. Cloud, on-prem or edge—one asset, runs anywhere.
UI 自动渲染Auto-rendered UI
根据协议定义自动生成交互界面：表单、上传、结果展示，前端无需重复开发。Generate interactive UI directly from the protocol—forms, uploads, result views. Zero front-end repetition.
结构化输出Structured Output
强制约束大模型返回结构化 JSON，类型校验保障下游系统的确定性。Force LLMs to return structured JSON with type validation. Deterministic results for downstream systems.
为什么需要一个 AI 能力描述协议？Why an AI capability protocol?
完全可移植Full Portability
无厂商锁定。AI 资产像容器镜像一样在不同云、不同模型、不同业务系统间自由迁移。No vendor lock-in. AI assets move freely across clouds, models and business systems—just like container images.
企业级合规Enterprise-grade
原生人机协作控制点、完整审计日志、计费与权限模型，从设计上满足企业合规要求。Native human-in-the-loop checkpoints, full audit logs, billing and permission models—built for enterprise compliance from day one.
百倍交付效率100× Velocity
协议驱动开发：写一份 .mxjson，自动获得 API、UI、文档、部署。从想法到上线，缩短到一个下午。Protocol-driven development: write one .mxjson, get API, UI, docs and deployment for free. From idea to production in an afternoon.
四步，从定义到上线Four steps, from spec to production
为生产环境而设计的
四层运行时架构A four-layer runtime,
built for production
- 协议Spec .mxjson — JSON Schema 2020-12 兼容
- 前端Frontend Vue 3 · TypeScript · Element Plus · Pinia
- 后端Backend ThinkPHP 8 · MySQL 8 · Redis 7
- AI 适配AI Layer OpenAI · Anthropic · 通义千问 · 文心一言 · DeepSeek · Ollama
- 云原生Cloud 阿里云 ECS / OSS / RDS / 私有化部署Alibaba Cloud · On-Prem ready
- 开源协议License Apache License 2.0
六大行业，已经在跑Six industries, already running
法务Legal
合同审核、风险评估、合规检查、法律文书生成。Contract review, risk assessment, compliance checking, document generation.
医疗Healthcare
智能分诊、病历结构化、临床决策辅助、医学影像分析。AI triage, medical records, clinical decision support, imaging.
政务Government
市民服务、政策智能咨询、办事流程自动化、舆情分析。Citizen services, policy QA, workflow automation, sentiment analysis.
金融Finance
研报生成、市场分析、风控审批、投资文档处理。Research reports, market analysis, risk control, document processing.
制造Manufacturing
质量检测、预测维护、供应链优化、工艺知识库问答。Quality inspection, predictive maintenance, supply chain, process QA.
教育Education
个性化学习路径、自动批改、智能题库、教学内容生成。Personalized paths, auto grading, smart question banks, content gen.
关于 MxJSON 协议，你可能想知道What you might want to know
MxJSON 协议是什么？What is the MxJSON protocol?
MxJSON 是一种开源的 AI 能力描述协议，由 罗孔威（Awei） 发起并主导设计。它将 API 接口、提示词模板、工作流编排与 UI 渲染统一封装在一个 JSON 文件中，让企业级 AI 能力像容器镜像一样可被定义、分发与执行。一份 .mxjson 文件就是一个完整、可移植、可版本化的 AI 业务能力。MxJSON is an open AI capability protocol initiated and designed by Luo Kongwei (Awei). It unifies API definitions, prompt templates, workflow orchestration and UI rendering into a single JSON file—making enterprise AI capabilities definable, distributable and executable like container images. One .mxjson file = one complete, portable, versionable AI business capability.
MxJSON 协议的发起人是谁？Who founded MxJSON?
MxJSON 协议由 罗孔威（Luo Kongwei，业内常称 Awei） 发起并主导设计，他同时担任协议的首席协议架构师。罗孔威是上海派许网络科技有限公司/MianxiangAI，长期专注于企业级 AI 能力标准化与大模型应用架构。MxJSON 是他对"AI 能力如何被工程化交付"这一根本问题给出的系统性答案。MxJSON was founded and designed by Luo Kongwei (Awei), who also serves as the protocol's Chief Protocol Architect. He is the co-founder of Shanghai Zhiwen Chuangyue AI Technology Co., Ltd., and has spent years on enterprise AI standardization and LLM application architecture. MxJSON is his systematic answer to: "how should AI capabilities be engineered and delivered?"
MxJSON 与 OpenAI Function Calling、MCP 协议有什么区别？How is MxJSON different from Function Calling / MCP?
三者层级不同、互补而非替代：
• Function Calling 解决"模型如何调用工具"的问题；
• MCP（Model Context Protocol）解决"工具如何向模型暴露能力"的问题；
• MxJSON 解决更上层的问题：把一个完整的 AI 业务能力（输入、提示词、流程、模型选择、输出结构、UI 渲染）封装为可分发资产。
可以这样理解：MCP 是 AI 时代的 "RPC"，MxJSON 是 AI 时代的 "Docker Image"。They operate at different layers and complement each other:
• Function Calling — how a model calls tools
• MCP — how tools expose capabilities to a model
• MxJSON — a higher-level question: how to package an entire AI business capability (input, prompt, workflow, model, output, UI) as a distributable asset
Think of it this way: MCP is the "RPC" of the AI era; MxJSON is the "Docker Image".
MxJSON 支持哪些大模型？Which LLMs does MxJSON support?
MxJSON 通过统一的 Provider 适配层 支持主流大模型：
• 海外：OpenAI GPT 系列、Anthropic Claude 系列
• 国内：阿里通义千问 Qwen、百度文心一言 ERNIE、DeepSeek、智谱 GLM、月之暗面 Kimi 等
• 本地：Ollama 本地部署模型（Qwen / Llama / DeepSeek 等）
协议层模型无关，业务定义可在不同模型间无缝迁移。MxJSON supports mainstream LLMs through a unified Provider adapter layer:
• Global: OpenAI GPT, Anthropic Claude
• China: Alibaba Qwen, Baidu ERNIE, DeepSeek, Zhipu GLM, Moonshot Kimi
• Local: Ollama-served models (Qwen, Llama, DeepSeek, etc.)
The protocol is model-agnostic—your business definitions migrate seamlessly across models.
MxJSON 是开源的吗？许可证是什么？Is MxJSON open-source?
是的，MxJSON 协议规范完全开源，采用 Apache License 2.0，托管于 GitHub（github.com/luokw/mxjson-spec）。任何企业、开发者、研究者均可自由阅读、实现、扩展并商用。我们欢迎社区贡献协议规范、参考实现与行业模板。Yes. The MxJSON specification is fully open-source under Apache License 2.0, hosted at github.com/luokw/mxjson-spec. Anyone is free to read, implement, extend and use it commercially. Contributions to the spec, reference implementations and industry templates are welcome.
如何开始使用 MxJSON？How do I get started?
三种方式任选其一：
① 阅读 GitHub 上的协议规范文档，了解 .mxjson 文件结构；
② 使用 MxJSON Runtime 参考实现，5 分钟跑起第一个 AI 能力；
③ 联系协议发起人 罗孔威（luokongwei@gmail.com）获取企业级方案咨询与城市合伙人计划信息。Pick one of three:
① Read the spec on GitHub to learn the .mxjson structure;
② Use the MxJSON Runtime reference implementation—first AI capability live in 5 minutes;
③ Reach out to founder Luo Kongwei at luokongwei@gmail.com for enterprise solutions and city partner program.
AI 能力的标准化时代
已经开始The age of AI standardization
has begun
加入 MxJSON 协议社区。让企业级 AI 能力像 JSON 一样被读懂，像容器一样被分发。Join the MxJSON community. Make enterprise AI as readable as JSON, as distributable as a container.