AI-Worker is a voice-first desktop agent workspace built on the Model Context Protocol (MCP). A unified hub for chat, files, browser automation, voice commands, and local-first LLM workflows — everything stays on your machine and under your control.
Most AI tools stop at chat. AI-Worker is built around the next step: connecting the assistant to useful local capabilities while keeping the workspace understandable, inspectable, and desktop-native.
Unlike browser-based AI tools or cloud-dependent agents:
- 🎤 Voice-First — Push-to-talk speech input with offline recognition (Vosk). Hands-free, privacy-first interactions.
- 🏠 Desktop-Native — Built with Electron. Works offline, integrates with your local files, no API dependency.
- 🔌 Extensible via MCP — Model Context Protocol integration for memory, filesystem, browser automation, and custom tools.
- 🌐 Local AI Ready — On-device model execution with WebGPU acceleration. Bring your own LLM (OpenAI, Gemini, Ollama, OpenRouter).
- ⚡ Agent Orchestration — Coordinate research, document extraction, file work, browser tasks, and WhatsApp workflows from one app.
- 🔒 Security & Control — Electron sandboxing, validated IPC, no external services required.
- Text and voice input (speech-to-text with Vosk)
- File drag-and-drop for context
- Workflow starter tiles for common tasks
- Rich conversation history with semantic search
- Built-in MCP clients for stdio and SSE connections
- Pre-configured servers: memory, filesystem, MarkItDown, browser automation
- Easy integration of custom MCP servers
- Real-time connection status and debugging
- OpenAI — GPT-4, GPT-3.5
- Gemini — Full API support
- Ollama — Local model hosting
- OpenRouter — 100+ models on one API
- Web-LLM — On-device WebGPU execution
- Auto-mode for intelligent provider routing
- Push-to-talk microphone — Click to record, auto-detect end
- Offline Vosk — Private speech recognition, no cloud dependency
- Voice command routing — Intent detection for hands-free workflows
- Playwright-backed — Full web automation (click, type, extract, screenshot)
- Secure sandboxing — Isolated browser context
- Form filling & navigation — Automate data entry and research workflows
- Direct phone messaging via Baileys library
- Approval workflows ("ask for permission via WhatsApp")
- Multi-user task delegation
- macOS — Universal (Intel + Apple Silicon)
- Linux — AppImage and deb support
- Windows — EXE installer
AI-Worker is engineered around three foundational pillars for modern agentic systems:
┌─────────────────────────────────────────────────────────────────┐
│ AI-Worker Architecture │
├───────────────────┬──────────────────────┬──────────────────────┤
│ Prompt Engineering │ Context Engineering │ Harness Engineering │
│ │ │ │
│ ∙ System Intent │ ∙ Semantic Memory │ ∙ Browser Sandboxing │
│ ∙ On-Device LLMs │ ∙ Context Pruning │ ∙ Secure IPC Bridges │
│ ∙ Tool Validation │ ∙ MCP Server Binding │ ∙ WhatsApp Gateway │
└───────────────────┴──────────────────────┴──────────────────────┘
- System Instruction Alignment — Structured prompts enforcing tool-use validation
- On-Device Inference — WebGPU-accelerated local model execution
- Dynamic Templates — Real-time variable injection for user directives and tool outputs
- Semantic Memory — Persistent entity & relationship database (MCP server-backed)
- Context Pruning — Algorithmic token reduction preserving workspace metadata
- Multi-Source Fusion — Unified context from files, browser state, terminal sessions, and memory
- Playwright Browser Harness — Secure web automation for extraction, forms, navigation
- Process Isolation — Electron sandbox + validated IPC boundaries
- Communication Bridges — Baileys/WhatsApp integration for approval workflows
- Node.js ≥ 22.12.0 — Install Node.js
- npm — Included with Node.js
- Python 3 + uv — For Python-based MCP servers (optional but recommended)
1. Clone the repository
git clone https://github.com/meharajM/ai-worker.app.git
cd ai-worker.app2. Install dependencies
npm install3. Bootstrap MCP and system dependencies
For macOS & Linux:
chmod +x ./scripts/setup-dependencies.sh
./scripts/setup-dependencies.shFor Windows (run PowerShell as Administrator):
.\\scripts\\setup-dependencies.ps14. Start development server
npm run devThe app opens at http://localhost:5173 with hot-reload. If Electron shell fails, the renderer is still available at that address.
5. Build native installers (optional)
# macOS (Intel + Apple Silicon)
npm run build:mac
# Linux
npm run build:linux
# Windows
npm run build:winFor detailed setup help, see Setup Guide.
- Start in Hub Chat — Type or use voice input
- Go to Hub Settings → LLM Providers → Choose your model source (OpenAI, Ollama, etc.)
- Configure MCP Connections — Inspect built-in tools or add custom servers
- (Optional) Enable Speech Settings → Configure Vosk or cloud speech services
The default local setup includes:
- Internal semantic memory (local SQLite)
- Filesystem access and watching
- MarkItDown document parsing
- Playwright-backed browser automation
- Setup Guide — Environment configuration, dependencies, troubleshooting
- Usage Guide — Voice controls, provider setup, MCP configuration, workflows
- MCP Integration — Build and connect custom MCP servers
- Docs Index — Full documentation map
# Development
npm run dev # Start dev server with hot-reload (HMR)
npm run dev:clean # Clear cache and cold-start dev
# Quality & Validation
npm run lint # Check code style (ESLint)
npm run lint:fix # Auto-fix linting issues
npm run typecheck # Verify TypeScript compilation
npm run build # Build Electron app
npm run test:mock # Run mocked E2E tests
# Production
npm run build:mac # Build macOS universal binary
npm run build:linux # Build Linux AppImage
npm run build:win # Build Windows installer
npm run publish:all # Publish to Cloudflare R2For full command docs, see Scripts README.
ai-worker.app/
├── src/
│ ├── main/ # Electron backend, native drivers, SQLite
│ ├── preload/ # IPC security bridges (isolated context)
│ ├── renderer/ # React UI and local LLM controllers
│ └── renderer/src/lib/agent/ # Intent analysis, orchestration, tool execution
├── docs/ # Guides, architecture, screenshots, proposals
├── tests/ # Playwright E2E tests, mocks, validation
├── scripts/ # Platform-specific setup and build scripts
└── build/ # Package assets (icons, splashscreens)
We love contributions! Start with CONTRIBUTING.md to understand our process.
- Bug Report — Found an issue?
- Feature Request — Have an idea?
- Security Policy — Report vulnerabilities responsibly
- Code of Conduct — Community standards
- Contribute code — See good-first-issues or open a discussion
- Improve docs — Setup guides, tutorials, examples
- Build MCP servers — Extend capabilities via protocol
- Provider integrations — Add new LLM providers or speech backends
- Share workflows — Post use cases in Discussions
AI-Worker is early-stage open-source software.
✅ **Production-ready:**
- Electron app with native packaging (macOS, Linux, Windows)
- React renderer UI with hot-reload
- MCP connection surface and configuration
- LLM provider settings and routing
- Speech recognition setup
- Browser automation via Playwright
- Release and publishing scripts
- Hardening for first-time contributors
- Dependency audit and security triaging
- GitHub repository settings review
Known: If Electron shell fails during development, the renderer is available at http://localhost:5173.
Help other builders discover AI-Worker:
- ⭐ Star the repository — Boosts GitHub search ranking
- 🔗 Share — Link to this repo with developers working on AI agents, browser automation, voice interfaces, or local-first workflows
- 💬 Join Discussions — Share your workflows, questions, and examples
- 🐛 Open Issues — Report bugs or request features
- 📝 Contribute — Code, docs, examples, integrations
AI-Worker is open-source software licensed under the MIT License.
- Homepage: https://ai-worker.tech
- GitHub: https://github.com/meharajM/ai-worker.app
- Author: AI-Worker Team team@aiworker.app
Built with ❤️ for developers and power users who want to reclaim control of their AI workflows.