An AI-native engineering intelligence platform powered by H-JEPA and Energy-Based Models
Features • Architecture • Quickstart • Docs • Contributing • License
UEWM (Universal Engineering World Model) is an AI system that builds a predictive world model of the entire software engineering lifecycle — from market analysis to production monitoring. Unlike traditional AI coding assistants that generate code token-by-token, UEWM reasons in a latent embedding space using Joint Embedding Predictive Architecture (H-JEPA) and makes decisions through Energy-Based Model (EBM) optimization.
Think of it as an AI that doesn't just write code — it understands your entire engineering organization and can predict the downstream consequences of every decision.
For detailed information on UEWM's core capabilities, please refer to the specific design documents:
- H-JEPA Brain Core & Architecture
- 12 Specialized AI Agents
- Self-Evolution Mechanisms
- Enterprise Security & Governance
- EIP Protocol Specifications
- Data & Privacy Strategy
Please refer to the comprehensive System Architecture Design document for detailed architectural diagrams, Brain Core component breakdowns, and sequence interactions.
⚠️ UEWM is currently in the Design Complete phase. Source code implementation begins in Phase 0. Star this repo to follow progress!
# Clone the repository
git clone https://github.com/YOUR_ORG/uewm.git
cd uewm
# (Coming in Phase 0) Setup development environment
# pip install -e ".[dev]"
# uewm init --profile s| Phase | Status | Description |
|---|---|---|
| Design | ✅ Complete | 10 design docs, 112 acceptance criteria, 100% coverage |
| Phase 0 | 🔄 Starting | MVLS (3 Z-Layers), inner ring 5 Agents, EIP protocol |
| Phase 1 | ⏳ Planned | +2 Z-Layers, middle ring Agents, multi-tenant, federated learning |
| Phase 2 | ⏳ Planned | Full 8 Z-Layers, outer ring Agents, SOC 2 Type II |
| Phase 3 | ⏳ Planned | Full self-optimization, all TRL-5 |
📖 Full English documentation → | 📖 完整中文文档 →
| Core Document | Path | Description |
|---|---|---|
| Requirements V6.1 | requirements/UEWM_Requirements_V6.1.md |
The unified source of truth for all requirements and ACs |
| System Architecture 1.1 | design/UEWM_Architecture_deliver_v1.1.md |
Core system architecture, Brain Core components, H-JEPA |
| Agents Design 1.1 | design/UEWM_Agents_Design_deliver_v1.1.md |
12 AI Agents, three-ring architecture, ALFA framework |
| Self-Evolution 1.1 | design/UEWM_Self_Evolution_deliver_v1.1.md |
Continuous learning, circuit breakers, Pareto optimization |
| EIP Protocol 1.1 | design/UEWM_EIP_Protocol_deliver_v1.1.md |
Engineering Intelligence Protocol, RPC/Messaging schemas |
| Safety Governance 1.1 | design/UEWM_Safety_Governance_deliver_v1.1.md |
Security framework, RBAC compliance, SOC 2 roadmap |
| Data Strategy 1.1 | design/UEWM_Data_Strategy_deliver_v1.1.md |
Training pipelines, knowledge transfer, federated learning |
| Deployment Ops 1.1 | design/UEWM_Deployment_Operations_deliver_v1.1.md |
Cluster topology, observability, error budget policies |
| Integration Map 1.1 | design/UEWM_Integration_Map_deliver_v1.1.md |
External tool adapters, system boundaries layer |
| Engineering Spec 1.1 | design/UEWM_Engineering_Spec_deliver_v1.1.md |
API dependencies, sequence architectures |
| Traceability Matrix 1.1 | design/UEWM_Traceability_Matrix_deliver_v1.1.md |
End-to-end design requirement coverage assessment |
Note: Historic architecture iterations, coverage reports, and GAP analyses are maintained in the
docs/en/design/arch_20260324anddocs/en/design/arch_20260322archives.
We welcome contributions! Please read our Contributing Guide and sign our CLA before submitting PRs.
Good first issues:
- Implement EIP Protobuf IDL compilation pipeline
- Create Z-Buffer Manager basic read/write operations
- Build Agent state machine framework
- Set up CI/CD pipeline with GitHub Actions
UEWM is dual-licensed:
- Open Source: GNU Affero General Public License v3.0 — free to use, modify, and deploy. If you offer UEWM as a network service, your modifications must be open-sourced under AGPL.
- Commercial: Contact us for a commercial license if AGPL doesn't fit your use case.
Note: Pre-trained model weights and customer-specific LoRA adapters are NOT covered by the AGPL license and are proprietary.
UEWM（通用工程世界模型）是一个 AI 系统，它为整个软件工程生命周期（从市场分析到生产监控）构建预测性世界模型。与传统逐 token 生成代码的 AI 编程助手不同，UEWM 在隐空间嵌入中使用联合嵌入预测架构（H-JEPA）进行推理，并通过**能量基模型（EBM）**优化做出决策。
它不只是写代码——它理解你的整个工程组织，并能预测每个决策的下游影响。
关于 UEWM 的详细核心能力与系统架构图纸，请参见对应的独立设计文档：
📖 完整中文文档目录 → | 📖 Full English documentation →
| 核心交付文档 | 相对路径 | 文档说明 |
|---|---|---|
| 需求规格 V6.1 | requirements/UEWM_Requirements_V6.1.md |
全项目需求与验收指标 (AC) 的唯一基线引用源 |
| 系统架构 1.1 | design/UEWM_Architecture_deliver_v1.1.md |
系统整体架构、Brain Core 组件、H-JEPA 底座 |
| 智能体设计 1.1 | design/UEWM_Agents_Design_deliver_v1.1.md |
12 个工程 AI Agent、三环架构、ALFA 控制框架 |
| 自进化机制 1.1 | design/UEWM_Self_Evolution_deliver_v1.1.md |
持续学习、安全断路器、帕累托进化优化验证 |
| EIP 协议 1.1 | design/UEWM_EIP_Protocol_deliver_v1.1.md |
各组件通信 RPC 与消息强类型载荷规范 |
| 安全治理 1.1 | design/UEWM_Safety_Governance_deliver_v1.1.md |
访问控制框架、RBAC 矩阵、SOC 2 安全演练路线 |
| 数据策略 1.1 | design/UEWM_Data_Strategy_deliver_v1.1.md |
训练流水线、基础模型选型与知识联邦学习策略 |
| 部署运维 1.1 | design/UEWM_Deployment_Operations_deliver_v1.1.md |
拓扑图、可观测性规划、SLO 与错误预算响应 |
| 集成边界 1.1 | design/UEWM_Integration_Map_deliver_v1.1.md |
外部工具适配器层、内外置模型调用系统边界 |
| 工程规格 1.1 | design/UEWM_Engineering_Spec_deliver_v1.1.md |
组件接口调用序列结构细节与配置依赖表 |
| 追溯矩阵 1.1 | design/UEWM_Traceability_Matrix_deliver_v1.1.md |
需求到具体系统架构环节的 100% 测试覆盖映射 |
注：其余各版本的迭代架构合并记录、差异补丁包和覆盖率分析文档，请分别查阅
docs/zh/design/arch_20260324及docs/zh/design/arch_20260322存档目录。
UEWM 采用双许可：
注意： 预训练模型权重和客户专属 LoRA 适配器不在 AGPL 许可范围内，属于专有资产。
Built with ❤️ by the UEWM Team