7-Brain Reasoning · 40+ Agent Types · 14 Swarm Patterns · Multi-Model Failover · Tauri v2 Desktop
Install · Architecture · Features · API · Uninstall
Agentic OS X is a kernel-first autonomous AI operating system designed to be lightweight as Claude Desktop, responsive as Cursor, autonomous as Hermes, and extensible as OpenHands. It runs natively on Windows via Tauri v2 with a bundled Node.js runtime — no external dependencies required.
The core kernel contains only 9 runtimes (Agent, Swarm, Brain, Memory, Knowledge, Model Router, Event Bus, Security, Observability). Everything else — agents, swarms, brains, providers — is lazy-loaded on demand and unloaded when idle. No bloated startup. No resident processes. No wasted resources.
Download the latest AgenticOS-Setup-2.0.0.exe from the Releases page and run it. The installer:
- Bundles Node.js v20 LTS — no Node.js installation required
- Installs to
%LOCALAPPDATA%\Agentic OS - Creates Start Menu shortcuts and desktop icon
- Registers uninstaller in Add/Remove Programs
- Auto-starts the application after installation
# Prerequisites: Node.js 20+, NSIS 3.x, Git
git clone https://github.com/rachidSabah/AgentincOS-Native.git
cd AgentincOS-Native
npm install
npm run build
powershell -File installer\build-exe.ps1The built installer will be in ./dist/AgenticOS-Setup-2.0.0.exe.
git clone https://github.com/rachidSabah/AgentincOS-Native.git
cd AgentincOS-Native
npm install
npm run dev # Development at http://localhost:3000
npm run build # Production build
npm start # Production server at http://localhost:3000npm run tauri:dev # Development with hot reload
npm run tauri:build # Production .msi/.exe bundle┌──────────────────────────────────────────────────────────────────┐
│ AGENTIC OS X KERNEL │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
│ │ Agent │ │ Swarm │ │ Brain │ │ Memory │ │
│ │ Runtime │ │ Runtime │ │ Runtime │ │ Runtime │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────────────┘ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
│ │Knowledge │ │ Model │ │ Event │ │ Security │ │
│ │ Runtime │ │ Router │ │ Bus │ │ Runtime │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────────────┘ │
│ ┌──────────┐ │
│ │Observab. │ ← Only 9 core runtimes. Everything else is │
│ │ Runtime │ loaded on-demand and unloaded when idle. │
│ └──────────┘ │
│ │
│ Lazy-Loaded Modules (NOT resident at startup): │
│ ┌─────────────────────────────────────────────────────────┐ │
│ │ 40+ Agent Types │ 14 Swarm Patterns │ 7 Brain Modes │ │
│ │ 10+ AI Providers │ Memory Pool │ Knowledge Graph │ │
│ └─────────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────────┘
| Brain | Role | When Activated |
|---|---|---|
| Planning | Goal decomposition, task sequencing | Complex multi-step tasks |
| Reasoning | Logical inference, deduction | Analysis and problem-solving |
| Architecture | System design, structural decisions | Code and system design |
| Research | Information gathering, synthesis | Research and exploration |
| Coding | Code generation, debugging | Development tasks |
| Memory | Context recall, knowledge retrieval | Memory-dependent tasks |
| Verification | Output validation, quality checks | Review and testing |
Request → Task Classifier → Provider Selection → Circuit Breaker → Execute
↓
Failover Chain (if primary fails):
Gemini CLI → Gemini API → Claude → GLM → Mistral → OpenAI
↓
Circuit Breaker States: Closed → Half-Open → Open
Provider Pooling: Lazy-init, connection reuse, idle unload
- 10+ Providers: Gemini CLI, Gemini API, OpenAI, Claude, GLM, Mistral, DeepSeek, OpenRouter, Qwen, Grok, Moonshot, Ollama, LM Studio, vLLM, llama.cpp
- Native SDK calls — no gateway middleware. Each provider called directly via its own SDK.
- Traffic shaping — coding tasks → DeepSeek, research → Gemini, planning → Claude
- Circuit breakers — auto-trip on failures, half-open recovery, per-provider isolation
- Idle unloading — providers unused for 5 minutes are automatically unloaded
Agents and swarms are never loaded at startup. They are created when needed and destroyed after completion.
Agent Library (40+ types across 7 categories):
| Category | Agent Types |
|---|---|
| Executive | CEO, COO, CFO, Strategist |
| Engineering | Frontend, Backend, DevOps, QA, Architect |
| Research | Researcher, Analyst, Librarian, Fact-Checker |
| Business | Marketer, Sales, Recruiter, Product Manager |
| Data | Data Engineer, ML Engineer, Analyst, Visualization |
| Knowledge | Knowledge Manager, Curator, Ontologist |
| Security | Security Auditor, Pen Tester, Compliance Officer |
Swarm Library (14 patterns):
| Swarm | Purpose |
|---|---|
| Software Factory | Full-stack development with specialized agents |
| Research | Parallel research with synthesis |
| Marketing | Campaign creation and analysis |
| Recruitment | Candidate screening and evaluation |
| Security | Vulnerability scanning and remediation |
| Data | ETL pipeline and analytics |
| Knowledge | Knowledge base construction |
| DevOps | CI/CD and infrastructure automation |
| Cloud Infrastructure | Cloud deployment and management |
| WordPress | Site building and content management |
| Moodle | Course creation and LMS management |
| Aviation Recruitment | Specialized aviation hiring |
| Customer Success | Support and retention |
| Executive Planning | Strategic planning and decision-making |
| Feature | Description |
|---|---|
| Kernel-First Architecture | Only 9 core runtimes — everything else is modular and on-demand |
| Lazy Loading | Agents, swarms, and providers loaded only when requested |
| Idle Unloading | Unused providers and agents automatically cleaned up after 5 min |
| Circuit Breakers | Per-provider fault isolation with half-open recovery |
| Cold Start <3s | Dashboard loads in under 1 second |
| Feature | Description |
|---|---|
| 7-Brain Reasoning | Planning, Reasoning, Architecture, Research, Coding, Memory, Verification |
| Multi-Model Failover | 10+ providers with automatic failover chain |
| Gemini CLI Auto-Discovery | Detects local Gemini CLI, registers models dynamically |
| Task-Aware Routing | Coding → DeepSeek, Research → Gemini, Planning → Claude |
| Swarm Intelligence | 14 prebuilt swarm patterns for complex multi-agent tasks |
| Feature | Description |
|---|---|
| Shared Memory Pool | No duplicate storage across agents |
| Memory Compression | Context and artifact compression for efficiency |
| Session Snapshotting | Full session state persistence |
| Incremental Persistence | Only changed data is written |
| 9 Memory Types | Long-term, Short-term, Episodic, Semantic, Context, Project, User, Team, Conversation |
| Feature | Description |
|---|---|
| Tauri v2 Desktop | Native Windows app with Rust backend |
| NSIS One-Click Installer | Bundled Node.js v20 — zero external dependencies |
| Start Menu Integration | Shortcuts, desktop icon, Add/Remove Programs |
| Clean Uninstaller | Complete removal with registry cleanup |
| Feature | Description |
|---|---|
| RBAC | Role-based access control with permission scopes |
| Audit Trail | Complete activity logging and compliance |
| Encryption | AES-256-GCM encryption for sensitive data |
| Observability | Traces, metrics, logs with OpenTelemetry-compatible spans |
| Self-Healing | Automatic recovery from common failure modes |
| Route | Purpose |
|---|---|
/api/kernel |
Kernel status, runtime management |
/api/agents |
Agent lifecycle (spawn, execute, terminate) |
/api/swarm |
Swarm orchestration (create, deploy, dissolve) |
/api/models |
Model router, provider health, circuit breakers |
/api/gemini-cli |
Gemini CLI discovery, model registration |
/api/memory |
Memory CRUD, search, compression |
/api/knowledge |
Knowledge graph, RAG queries |
/api/knowledge/search |
Vector + graph + keyword search |
/api/knowledge/rag |
Retrieval-Augmented Generation |
/api/chat |
Multi-model chat with failover |
/api/browser |
Navigate, extract, analyze, search, summarize |
/api/auth |
Authentication and session management |
/api/audit |
Audit trail queries |
/api/observability |
Metrics, traces, logs |
/api/self-healing |
Health checks and auto-recovery |
/api/readiness |
System readiness validation |
/api/terminal |
Terminal command execution |
/api/artifacts |
Artifact management and preview |
Use Add/Remove Programs in Windows Settings, or run:
"%LOCALAPPDATA%\Agentic OS\uninstall.bat"Remove-Item -Recurse -Force "$env:LOCALAPPDATA\Agentic OS"curl -fsSL https://raw.githubusercontent.com/rachidSabah/AgentincOS-Native/main/uninstall.sh | bashCreate a .env file in the project root:
# ─── AI Provider Keys (only set the ones you use) ───
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GEMINI_API_KEY=AI...
DEEPSEEK_API_KEY=sk-...
OPENROUTER_API_KEY=sk-or-...
GLM_API_KEY=...
MISTRAL_API_KEY=...
QWEN_API_KEY=...
XAI_API_KEY=...
MOONSHOT_API_KEY=...
# ─── Local Providers (auto-detected) ───
# Ollama: http://localhost:11434
# LM Studio: http://localhost:1234
# vLLM: http://localhost:8000
# ─── Gemini CLI (auto-discovered) ───
# No API key needed — uses local `gemini` CLI if available
# ─── Database ───
DATABASE_URL=file:./agentic-os.db
# ─── Auth ───
NEXTAUTH_SECRET=your-secret-here
NEXTAUTH_URL=http://localhost:3000AgentincOS-Native/
├── installer/ # NSIS Windows installer
│ ├── installer-oneclick.nsi # One-click installer script
│ ├── installer.nsi # Full installer script
│ ├── build-exe.ps1 # PowerShell installer builder
│ ├── build-exe.sh # Linux/macOS builder
│ ├── stage.js # Installer staging script
│ ├── icon.ico # Application icon
│ ├── start.bat # Windows start script
│ ├── stop.bat # Windows stop script
│ └── uninstall.bat # Windows uninstall script
├── src-tauri/ # Tauri v2 desktop shell (Rust)
│ ├── src/ # Rust modules, plugins, commands
│ ├── Cargo.toml # Rust dependencies
│ └── tauri.conf.json # Tauri configuration
├── src/
│ ├── app/
│ │ ├── page.tsx # Main dashboard
│ │ ├── layout.tsx # Root layout
│ │ ├── globals.css # Cyberpunk theme
│ │ └── api/ # 18 API routes
│ ├── components/ # 16+ UI components
│ │ ├── ui/ # shadcn/ui primitives
│ │ └── ... # Dashboard, agents, swarms, etc.
│ ├── lib/
│ │ ├── kernel.ts # Core kernel with 9 runtimes
│ │ ├── model-router.ts # Multi-model router with circuit breakers
│ │ ├── agent-runtime.ts # Lazy-loaded agent system
│ │ ├── swarm-engine.ts # On-demand swarm orchestration
│ │ ├── memory-optimizer.ts # Memory pool, compression, snapshotting
│ │ ├── gemini-cli-discovery.ts # Auto-discovery engine
│ │ ├── readiness-validator.ts # System readiness checks
│ │ ├── types.ts # TypeScript type definitions
│ │ └── db.ts # Prisma database client
│ └── hooks/ # React hooks
├── prisma/
│ └── schema.prisma # 18 database models
├── scripts/ # Utility scripts
├── uninstall.sh # Linux/macOS uninstaller
├── uninstall.ps1 # Windows PowerShell uninstaller
├── next.config.ts # Next.js configuration
├── tailwind.config.ts # Tailwind configuration
└── package.json # Dependencies & scripts
| Metric | Target | Status |
|---|---|---|
| Cold Start | <3s | Achieved |
| Dashboard Load | <1s | Achieved |
| Agent Spawn | <500ms | Achieved |
| Swarm Spawn | <2s | Achieved |
| Memory Query | <100ms | Achieved |
| Artifact Preview | <300ms | Achieved |
| API Response | <500ms | Achieved |
| Layer | Technology |
|---|---|
| Framework | Next.js 16 (App Router) |
| Desktop | Tauri v2 (Rust) |
| Language | TypeScript + Rust |
| Styling | Tailwind CSS 4 + Glassmorphism |
| Animation | Framer Motion |
| State | Zustand |
| UI Components | shadcn/ui + Radix |
| Database | Prisma ORM (SQLite) |
| AI Providers | OpenAI, Anthropic, Google GenAI (native SDKs) |
| Installer | NSIS 3.x (LZMA compressed) |
| Runtime | Node.js v20 LTS (bundled) |
- Windows 10/11 (for NSIS installer and Tauri desktop)
- Node.js 20+ (only for development — installer bundles Node.js)
- npm or bun package manager
- Git for cloning and updates
- NSIS 3.x (only for building the installer)
MIT License — see LICENSE for details.
Built by rachidSabah