Agentry is a cross-platform desktop application that lets you hire and manage AI employees locally. Each employee is powered by a Skill — a self-contained package of expertise, tools, and behavioral rules — and runs on your machine with your own API keys.
Skill = The employee's soul (expertise + tool capabilities)
Agentry = The employee's body (runtime + management UI + collaboration engine)
BYOK (Bring Your Own Key) — Agentry connects directly to LLM providers using your own API keys. Your data never passes through our servers.
Hire AI employees from the marketplace or create your own. Each employee has an independent chat session, persistent memory, and specialized tools.
- One-click activation — Install a Skill, activate an employee, start chatting
- State machine —
offline → idle → working → idle / blocked / error - Independent sessions — Each employee gets their own Gateway session with full context isolation
Describe your goal, and the Supervisor PM automatically orchestrates your team:
- Plan — Decomposes your goal into a task DAG with dependencies
- Execute — Dispatches tasks to employees in dependency-wave order
- Synthesize — Aggregates results into a final deliverable
A Kanban board tracking all work across your AI team:
pending → in_progress → in_review → completed / blocked
Tasks support priorities, dependencies, assignees, ratings, and automatic execution.
| Layer | Purpose | Scope |
|---|---|---|
| Episodic | Task experiences & lessons learned | Per-employee |
| Semantic | Brand knowledge & business context | Shared across all employees |
Memory is automatically injected into system prompts at compile time.
- Markdown rendering with syntax highlighting
- Streaming output with thinking blocks
- Tool call status tracking
- File attachments
- Multi-conversation management per employee
Connect your AI employees to external platforms — Telegram, Discord, WhatsApp, Feishu, and more.
Automate recurring work with cron expressions — daily, weekly, monthly, or custom intervals.
- API keys encrypted at rest via OS-level cryptography (Keychain / DPAPI / libsecret)
- Context isolation with IPC whitelist — no Node.js access in the renderer
- Behavioral prohibition rules (hard/soft) to constrain employee actions
Agentry ships with 7 ready-to-use employees:
| Employee | Role | Type | Tools |
|---|---|---|---|
| Supervisor | PM & Team Lead | knowledge | Orchestrates other employees |
| Researcher | Research Analyst | knowledge | Web search + LLM reasoning |
| New Media | Content Creator | hybrid | DeerAPI image generation |
| Browser Agent | Web Automation | execution | Built-in browser control |
| Reddit Nurture | Reddit Growth | execution | Camofox headless browser |
| Publisher — Xiaohongshu | 小红书 Publisher | execution | xiaohongshu-mcp service |
| Publisher — Douyin | 抖音 Publisher | execution | social-auto-upload + Playwright |
Install more employees from the Skill Marketplace, or create your own.
Download the latest release for your platform:
| Platform | Architecture | Download |
|---|---|---|
| macOS | Apple Silicon (M1/M2/M3/M4) | Agentry-*-mac-arm64.dmg |
| macOS | Intel | Agentry-*-mac-x64.dmg |
| Windows | x64 | Agentry-*-win-x64.exe |
| Windows | ARM64 | Agentry-*-win-arm64.exe |
| Linux | x64 | Agentry-*-linux-x86_64.AppImage |
| Linux | ARM64 | Agentry-*-linux-arm64.AppImage |
- Open Agentry — The setup wizard will guide you through initial configuration
- Add an AI Provider — Configure at least one LLM provider with your API key (OpenAI, Anthropic, Google, DashScope, etc.)
- Activate an Employee — Go to the Employee Hub, pick an employee, click ▶ to activate
- Start Chatting — Click on the employee to open a conversation
- Node.js 22+
- pnpm 10+
- Git
# Clone the repository
git clone https://github.com/Youhai020616/Agentry.git
cd Agentry
# Install dependencies + download bundled runtime
pnpm init
# Start development server
pnpm dev| Command | Description |
|---|---|
pnpm dev |
Start dev server (Vite + Electron) |
pnpm build |
Full production build + electron-builder |
pnpm build:vite |
Vite build only (no packaging) |
pnpm lint |
ESLint with auto-fix |
pnpm typecheck |
TypeScript strict check (zero errors required) |
pnpm test |
Run unit tests (Vitest) |
pnpm test:e2e |
Run E2E tests (Playwright) |
pnpm package:mac |
Package for macOS |
pnpm package:win |
Package for Windows |
pnpm package:linux |
Package for Linux |
# Full build (all platforms depend on your OS)
pnpm build
# Platform-specific
pnpm package:mac # macOS (.dmg)
pnpm package:win # Windows (.exe)
pnpm package:linux # Linux (.AppImage, .deb, .rpm)┌─ Renderer (React 19) ──────────────────────────────────┐
│ Pages → Zustand Stores → window.electron.ipcRenderer │
└────────────────────┬────────────────────────────────────┘
│ IPC (Preload whitelist)
┌────────────────────▼────────────────────────────────────┐
│ Main Process (Electron 40, Node 22) │
│ │
│ ┌────────────────────────────────────────────────────┐ │
│ │ Skill Runtime Engine │ │
│ │ EmployeeManager · SkillCompiler · ManifestParser │ │
│ │ TaskQueue · Supervisor · TaskExecutor │ │
│ │ MemoryEngine · MessageBus · ProhibitionEngine │ │
│ │ CreditsEngine · ToolRegistry · BrowserManager │ │
│ └────────────────────────────────────────────────────┘ │
│ │
│ IPC Handlers (178) · Tray · Auto-Updater │
└────────────────────┬─────────────────────────────────────┘
│ JSON-RPC 2.0 over WebSocket
▼
OpenClaw Gateway (:18790)
→ LLM Providers (BYOK)
| Layer | Technology |
|---|---|
| Shell | Electron 40 |
| UI | React 19 + Tailwind CSS 3.4 + shadcn/ui |
| Language | TypeScript 5.7 (strict) |
| Routing | React Router 7 |
| State | Zustand 5 |
| i18n | i18next (English, 中文, 日本語) |
| Database | better-sqlite3 (5 databases, WAL mode) |
| Gateway | OpenClaw (JSON-RPC 2.0) |
| Build | Vite 7 + electron-builder 26 |
| Test | Vitest 4 + Playwright |
| Animations | Framer Motion 12 |
Scaffold a new Skill package:
pnpm create-skill my-employeeThis creates the standard Skill structure:
resources/employees/my-employee/
├── manifest.json # Metadata, capabilities, tools, secrets, onboarding
├── SKILL.md # System prompt template (supports {{ROLE}}, {{TEAM}}, etc.)
├── scripts/ # Execution scripts (optional)
├── tools/ # CLI tool definitions (optional)
└── references/ # Knowledge base files (optional)
Employee types:
| Type | Description | Example |
|---|---|---|
knowledge |
Conversation-only, no external tools | Supervisor, Researcher |
execution |
Requires external tools to operate | Reddit Nurture, Publishers |
hybrid |
Conversation + tools | New Media Creator |
- AI Employee lifecycle management
- Supervisor multi-agent orchestration
- Task Board with automatic execution
- Memory system (episodic + semantic)
- Behavioral prohibition rules
- Chat with streaming, tool calls, file attachments
- Channel integrations (Telegram, Discord, WhatsApp, etc.)
- Scheduled tasks (cron)
- Auto-updater (Alibaba Cloud OSS + GitHub Releases)
- More built-in employees
- Skill Marketplace improvements
- Desktop ↔ Cloud data sync
- Gateway migration to ZeroClaw (Rust)
- Agentry Cloud — 7×24 AI employees in the cloud
Contributions are welcome! Please read the project conventions in CLAUDE.md before submitting a PR.
- Fork the repository
- Create your feature branch (
git checkout -b feature/amazing-feature) - Run checks before committing:
pnpm typecheck && pnpm lint && pnpm test
- Commit your changes (
git commit -m 'feat: add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
MIT © Agentry Team
Agentry — Your Desktop AI Workforce