AI-Native Software Development Lifecycle Platform
Give every team an AI-native R&D system — not just Big Tech.
VibeOS is an open-source platform that brings 8 specialized AI agents to cover your entire software development lifecycle — from requirements to monitoring. Unlike code completion tools (Copilot, Cursor) or single-task agents (Devin), VibeOS orchestrates the full SDLC with cross-project memory that makes agents smarter over time.
3 commands to a running AI R&D platform:
make install && make infra && make dev
# Open http://localhost:3000Early-stage notice: VibeOS is under active development. You may encounter rough edges and occasional bugs. But it's battle-tested enough to demonstrate what AI-native SDLC looks like — and to give your team a real head start on AI infrastructure transformation. Issues and PRs are welcome.
8 specialized AI agents spanning the entire delivery lifecycle:
PM Agent → Architecture → Requirements → Design → Development → Testing → CI/CD → Monitoring
While Copilot helps you write code faster, VibeOS helps your entire team ship products faster.
- Agents collaborate through a graph-based workflow engine
- Cross-workspace memory (Mem0 + Qdrant) — agents get smarter with every project
- Knowledge distillation turns execution history into reusable institutional knowledge via Apache AGE graph
| Provider | Models | Region |
|---|---|---|
| OpenAI | GPT-4o, etc. | Global |
| Anthropic | Claude Sonnet/Opus | Global |
| DeepSeek | DeepSeek-Chat/Coder | Global / China |
| Volcengine | Doubao series | China |
| Dashscope | Qwen series | China |
| Any OpenAI-compatible | Custom | Anywhere |
Built-in model routing with capability contracts, token budget control, and circuit breaker via llm-gateway.
- Go backend services (workspace-svc, ws-gateway) for performance
- Python agents and platform services for AI flexibility
- React 19 frontend with real-time WebSocket collaboration
- PostgreSQL + Redis + Qdrant — proven infrastructure stack
- Docker Compose one-command deployment
- coding-agent powered by OpenHands SDK — not just file snippets, full repository operations
- GitLab integration for real branch/MR workflows
- Guard hooks and stuck detection for safe autonomous coding
- Add custom agents, tools, and workflow phases
- Optional integrations: GitLab, Feishu/Lark, Tencent COS
- Apache 2.0 — use it, fork it, build on it
| Capability | VibeOS | GitHub Copilot | Cursor | Devin |
|---|---|---|---|---|
| Code completion | via coding-agent | Yes | Yes | Yes |
| Requirements analysis | Yes | - | - | - |
| Architecture design | Yes | - | - | - |
| Full SDLC workflow | Yes | - | - | Partial |
| Multi-agent orchestration | Yes | - | - | - |
| Cross-project memory | Yes | - | - | - |
| Knowledge graph | Yes | - | - | - |
| Self-hosted / open source | Yes | - | - | - |
| LLM provider choice | Any | OpenAI | Multiple | Proprietary |
┌──────────────────────────────────────────────────────────┐
│ Frontend (React 19) │
│ Vite 8 · TypeScript · Tailwind 4 │
│ localhost:3000 │
└────────┬──────────────────┬───────────────┬──────────────┘
│ /api/* │ /api/nlp, │ /ws
│ │ /api/workflow, │
│ │ /api/feedback │
▼ ▼ ▼
┌─────────────┐ ┌─────────────┐ ┌──────────────┐
│workspace-svc│ │ pm-agent │ │ ws-gateway │
│ (Go/Chi) │ │ (FastAPI) │ │ (Go/WS) │
│ :8010 │ │ :8040 │ │ :8020 │
└──────┬──────┘ └──────┬──────┘ └──────┬───────┘
│ │ │
│ ┌──────┴──────┐ │
│ │ Dispatcher │ │
│ └──────┬──────┘ │
│ ┌───────────┼───────────┐ │
│ ▼ ▼ ▼ │
│ ┌──────┐ ┌──────────┐ ┌──────┐ │
│ │arch │ │dev-agent │ │ ... │ │
│ │:8041 │ │ :8044 │ │agents│ │
│ └──────┘ └──────────┘ └──────┘ │
│ │ │
▼ ▼ ▼
┌──────────┐ ┌───────────────┐ ┌──────────────┐
│PostgreSQL│ │ llm-gateway │ │ Redis │
│ :5432 │ │ :8030 │ │ :6379 │
└──────────┘ └───────────────┘ └──────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
┌──────────┐ ┌────────────┐ ┌──────────────┐
│ memory- │ │rag-pipeline│ │ knowledge- │
│ service │ │ :8060 │ │ service │
│ :8050 │ └─────┬──────┘ │ :8070 │
└────┬─────┘ │ └──────┬───────┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────────┐
│ Qdrant │ │ Qdrant │ │ PostgreSQL │
│ :6333 │ │ :6333 │ │ (AGE graph) │
└──────────┘ └──────────┘ └──────────────┘
| Tool | Version | Install |
|---|---|---|
| Docker | Latest | docker.com |
| Go | 1.25+ | go.dev |
| Python | 3.12+ | python.org |
| uv | Latest | docs.astral.sh/uv |
| Node.js | 20+ | nodejs.org |
| pnpm | 9+ | npm install -g pnpm |
git clone https://github.com/jokeuncle/VibeOS.git
cd VibeOS
cp .env.example .envEdit .env and configure your LLM provider (at minimum, set one API key):
# Pick your provider — examples:
LLM_API_KEY=sk-xxxx # OpenAI
LLM_BASE_URL=https://api.openai.com/v1
LLM_MODEL=gpt-4o
# Or for DeepSeek:
# LLM_API_KEY=sk-xxxx
# LLM_BASE_URL=https://api.deepseek.com/v1
# LLM_MODEL=deepseek-chatSee .env.example for all provider options.
make install # Install all dependencies (Python + JS + Go)
make infra # Start Postgres, Redis, Qdrant via Docker
make db-init # Initialize database schema (first time only)
make db-migrate # Apply migrations
make dev # Start all services + frontendmake health # Check all service endpointsOpen http://localhost:3000 in your browser.
Run make help to see all available targets.
make run-workspace-svc # Go :8010
make run-ws-gateway # Go :8020
make run-llm-gateway # Py :8030
make run-memory-service # Py :8050
make run-rag-pipeline # Py :8060
make run-knowledge-service # Py :8070
make run-pm-agent # Py :8040
make run-architecture-agent # Py :8041
make run-dev-agent # Py :8044
make run-coding-agent # Py :8048
make run-web # JS :3000VibeOS routes all LLM calls through llm-gateway, which supports any OpenAI-compatible API. Configure via environment variables:
| Variable | Description | Example |
|---|---|---|
LLM_API_KEY |
API key for your LLM provider | sk-xxxx |
LLM_BASE_URL |
Base URL of the provider API | https://api.openai.com/v1 |
LLM_MODEL |
Default model name | gpt-4o |
For provider-specific keys (e.g. using multiple providers simultaneously), you can also set OPENAI_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY, VOLCENGINE_API_KEY individually.
| Integration | Required Env Vars | Purpose |
|---|---|---|
| GitLab | GITLAB_URL, GITLAB_TOKEN |
Code repository management, branch/MR workflows |
| Feishu/Lark | FEISHU_APP_ID, FEISHU_APP_SECRET |
Create Feishu docs from agent outputs |
| Tencent COS | VIBEOS_COS_UPLOAD=1, COS_* vars |
Upload artifacts to cloud storage |
| Service | Port | Role |
|---|---|---|
| Frontend (Vite) | 3000 | React SPA |
| workspace-svc | 8010 | REST API |
| ws-gateway | 8020 | WebSocket relay |
| llm-gateway | 8030 | LLM proxy |
| pm-agent | 8040 | Orchestrator |
| architecture-agent | 8041 | Architecture phase |
| requirement-agent | 8042 | Requirement phase |
| design-agent | 8043 | Design phase |
| dev-agent | 8044 | Development phase |
| test-agent | 8045 | Testing phase |
| cicd-agent | 8046 | CI/CD phase |
| monitoring-agent | 8047 | Monitoring phase |
| coding-agent | 8048 | Agentic coding |
| memory-service | 8050 | Memory (Mem0 + Qdrant) |
| rag-pipeline | 8060 | RAG indexing & retrieval |
| knowledge-service | 8070 | Knowledge graph |
| Layer | Technology | Purpose |
|---|---|---|
| Frontend | React 19, Vite 8, TypeScript, Tailwind 4, Zustand | SPA with real-time WebSocket updates |
| API Gateway | Go, Chi router | Workspace/task CRUD, auth |
| WebSocket | Go, gorilla/websocket, Redis Pub/Sub | Real-time event broadcasting |
| Agent Orchestrator | Python, FastAPI, LangGraph | Multi-agent workflow dispatch |
| Domain Agents | Python, FastAPI | Per-phase AI agents |
| LLM Gateway | Python, LiteLLM | Multi-provider routing, budget control |
| Memory | Python, Mem0, Qdrant | Four-layer memory system |
| RAG | Python, LlamaIndex, Qdrant | Per-workspace document indexing |
| Knowledge | Python, Apache AGE, PostgreSQL | Knowledge graph + distillation |
| Database | PostgreSQL 16 (with AGE extension) | Relational + graph data |
| Cache/PubSub | Redis 7 | Sessions, events, trust scores |
| Vector Store | Qdrant | Memory embeddings, RAG chunks |
VibeOS implements a multi-layered learning system where agents get smarter with every project:
Agent executes task
↓
┌─ add_memory() ─────────→ Mem0 (Project Memory, per workspace)
│
├─ _save_artifact() ──────→ RAG Pipeline (auto-indexed)
│ ↓
│ Qdrant (per-workspace collection)
│
├─ workflow:phase_complete → async _trigger_distill()
│ ↓
│ Knowledge Service (LLM extracts patterns)
│ ↓
│ AGE Graph (cross-workspace knowledge)
│
├─ workflow:task_complete ─→ async _auto_index_to_rag()
│
└─ user feedback (thumbs up/down) → PM Agent → Memory Service
↓
Preference Memory (improves future outputs)
| Layer | Scope | Storage | Purpose |
|---|---|---|---|
| L1 Working | Session | Mem0 (session_id) | Short-lived conversation context |
| L2 Project | Workspace | Mem0 (ws:{id}) | Tech stack, patterns, decisions |
| L3 Organization | Global | Mem0 (org:{id}) | Cross-workspace best practices |
| L4 Preference | Workspace | Mem0 (ws:{id}) | User feedback on agent outputs |
vibeos/
├── apps/web/ # React 19 SPA (pnpm workspace)
├── agents/ # Python AI agents (uv workspace)
│ ├── shared/vibeos_agent/ # Shared agent SDK
│ ├── pm-agent/ # Orchestrator (NLP + workflow)
│ ├── coding-agent/ # Agentic coding (OpenHands)
│ ├── architecture-agent/ # Architecture phase
│ ├── requirement-agent/ # Requirement phase
│ ├── design-agent/ # Design phase
│ ├── dev-agent/ # Development phase
│ ├── test-agent/ # Testing phase
│ ├── cicd-agent/ # CI/CD phase
│ └── monitoring-agent/ # Monitoring phase
├── services/ # Go backend services
│ ├── workspace-svc/ # REST API (Chi router)
│ ├── ws-gateway/ # WebSocket relay
│ └── shared/models/ # Shared DTOs
├── platform/ # Python platform services
│ ├── llm-gateway/ # Multi-provider LLM routing
│ ├── memory-service/ # Mem0 + Qdrant
│ ├── rag-pipeline/ # LlamaIndex + Qdrant
│ └── knowledge-service/ # Knowledge graph (AGE)
├── deploy/ # Docker, SQL, migrations
├── Makefile # Dev workflow automation
├── .env.example # Environment template
└── LICENSE # Apache 2.0
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
VibeOS is licensed under the Apache License 2.0.
If VibeOS helps your team, consider buying me a coffee. Your support keeps this project maintained and free for everyone.
No pressure — starring the repo is already a huge help!
VibeOS 是一个 AI 原生的软件开发生命周期平台，通过 8 个专业 AI Agent 协作完成从需求分析到持续监控的全流程。
与代码补全工具（Copilot、Cursor）或单任务 Agent（Devin）不同，VibeOS 编排 完整的 SDLC 流程，并具备跨项目的记忆积累能力，让 Agent 在反复执行任务的过程中持续变好。
- 全生命周期覆盖：PM → 架构 → 需求 → 设计 → 开发 → 测试 → CI/CD → 监控
- 多 Agent 协同：基于图的工作流引擎，Agent 间上下文自动传递
- 跨项目记忆：Mem0 + Qdrant 四层记忆体系，越用越聪明
- 知识蒸馏：Apache AGE 知识图谱，自动提炼组织级最佳实践
- LLM 无关：支持 OpenAI、Anthropic、DeepSeek、火山引擎等任意 OpenAI 兼容 API
- 生产级架构：Go + Python + React 19 微服务架构，Docker Compose 一键部署
cp .env.example .env # 配置 LLM API Key
make install # 安装依赖
make infra # 启动基础设施（Postgres、Redis、Qdrant）
make db-init # 初始化数据库（仅首次）
make db-migrate # 应用迁移
make dev # 启动所有服务
# 访问 http://localhost:3000- 想要搭上 AI Infra 转型列车的 中小型公司
- 希望在内部构建 AI 原生研发体系的 技术团队
- 对 AI Agent 工作流感兴趣的 开发者和研究者
VibeOS 仍在积极开发中，可能存在一些粗糙的边角。但它已经足够展示 AI 原生 SDLC 的样子，帮助你的团队抢先一步。欢迎提 Issue 和 PR！