An AI-powered intelligent browser built with Next.js and Electron. Features multi-modal AI task execution, scheduled tasks, social media integration, and advanced file management capabilities with support for multiple AI providers.
Built with Next.js and Electron.
⚠️ Important Notes
- This project is for learning purposes. There may be bugs and issues - your understanding is appreciated.
- Windows users: Building on Windows may require multiple attempts to succeed. If the build fails, try running the build command again.
- macOS recommended: The macOS version provides a better user experience. Limited resources have been allocated to Windows optimization.
- AI Provider note: Due to limited resources and network restrictions (no access to Claude, OpenAI, etc.), this project is primarily developed and tested with DeepSeek models. Other providers are supported but may have less thorough testing coverage. Community contributions for other providers are welcome!
- Frontend: Next.js 15 + React 19
- Desktop: Electron 33
- UI: Ant Design + Tailwind CSS
- State Management: Zustand
- Storage: IndexedDB (via electron-store)
- AI Agent: @jarvis-agent (based on Eko - production-ready agent framework)
- Build Tools: Vite + TypeScript
Node version: 20.19.3
First, run the development server:
# Install dependencies
pnpm install
# Build desktop application client for mac
pnpm run build:deps
# Build desktop application client for windows
pnpm run build:deps:win
# Start web development server
pnpm run next
# Start desktop application
pnpm run electron
After launching the application:
- Click the Settings icon (⚙️) in the top-right corner
- Navigate to Providers panel
- Select your AI provider (DeepSeek, Qwen, Google Gemini, Claude, or OpenRouter)
- Click Edit API Key and enter your API key
- Click the checkmark to save
For detailed configuration instructions, see CONFIGURATION.md.
To build the desktop application for distribution:
# Build the application for mac
pnpm run build
# Build the application for windows
pnpm run build:win
Note: End users will configure their API keys through the Settings UI after installation. No environment files needed.
- Dual Mode Interaction:
- Chat Mode (default): Lightweight conversation with ChatAgent — streaming text, tool calls (web search, webpage QA), thinking display
- Deep Explore Mode: Multi-step workflow generation → user confirm → multi-agent execution for complex tasks
- Web Search Integration: Built-in web search via Tavily, Serper, or SearXNG (self-hosted) with configurable search provider
- Multiple AI Providers: Support for DeepSeek, Qwen, Google Gemini, Anthropic Claude, and OpenRouter
- Complete Settings System: Configure everything through the UI - no file editing required
- Providers: API keys, models, and provider-specific settings
- General: Language, startup behavior, window preferences
- Chat: Temperature, max tokens, streaming, search provider configuration
- Agent: Browser/File/Custom agent settings, per-agent MCP services
- UI: Theme (Dark/Light/System), font size, density, editor preferences
- Network: Proxy, request timeout, stream timeout, retry attempts
- Workflow Control: Editable workflow steps with pause/resume, expert mode, and thinking display
- Scheduled Tasks: Create and manage automated recurring tasks with custom intervals
- AI-Powered Browser: Intelligent multi-tab browser with automated task execution
- Multi-Modal AI: Vision and text processing capabilities
- Speech & TTS: Voice recognition and text-to-speech integration
- File Management: Advanced file operations and management
- Internationalization: Full English and Chinese language support
v0.0.1 - v0.0.4: Core Functionality
- AI-powered browser with automated task execution
- Multiple AI provider support (DeepSeek, Qwen, Google Gemini, Claude, OpenRouter)
- Multi-modal AI capabilities (vision and text processing)
- Scheduled tasks system with custom intervals
- File management capabilities
- UI configuration for API keys and models
v0.0.5 - v0.0.7: UI/UX Enhancements
- Purple theme redesign with improved UI/UX
- Agent Configuration system (custom prompts, MCP tools management)
- Toolbox page (centralized feature hub)
- Internationalization support (English/Chinese)
- WebGL animated background with gradient fallback
- Improved modal sizes and layout optimization
v0.0.8 - v0.0.10: Advanced Features
- Human interaction support (AI can ask questions during execution)
- Task continuation with file attachment management
- Atomic fragment-based history playback with typewriter effects
- Advanced playback controls (play/pause/restart/speed adjustment)
- Context restoration and session management
- Optimized auto-scroll behavior for messages
- Enhanced message display and rendering
v0.0.11+: Unified Settings System
- Complete Settings redesign with 6 panels (Providers, General, Chat, Agent, UI, Network)
- Unified configuration management with electron-store
- All settings configurable through UI - no manual file editing required
- Real-time settings sync across all windows
- Import/Export/Reset functionality for all settings
- Dark/Light/System theme support with smooth transitions
- Network configuration (Proxy, Timeout, Retry)
- Scheduled Tasks system with IndexedDB storage
v0.0.12+: Chat Mode & Agent Enhancement
- Chat mode with ChatAgent (streaming text + inline tool calls)
- Deep Explore mode (renamed from workflow mode) for complex multi-step tasks
- Web search integration with 3 providers (Tavily, Serper, SearXNG)
- Webpage QA — extract and analyze content from open browser tabs
- Editable workflow steps with user confirmation before execution
- Task pause/resume support
- Expert mode with thinking process display
- Per-agent MCP service configuration
- Custom agent support with user-defined prompts
- Multi-window task isolation and global service injection
Phase 1: Enhanced User Experience
- Voice input support (speech-to-text integration)
- Theme customization system (multiple color schemes)
- Enhanced accessibility features
Phase 2: Workflow Enhancement
- Workflow configuration export/import functionality
- Refactored scheduled task steps based on workflow configuration
- Visual workflow editor with drag-and-drop interface
- Workflow templates and presets
Phase 3: Plugin Ecosystem
- MCP plugin marketplace
- Community plugin sharing platform
- Plugin version management system
- One-click plugin installation and updates
- Plugin development toolkit and documentation
Phase 4: Advanced Capabilities
- Collaborative task execution
- Cloud sync for tasks and configurations
- Mobile companion app
- Performance optimization and caching improvements
Animated splash screen on application launch.
Simple and clean interface - input your task and let AI execute automatically.
Left: AI thinking process and execution steps. Right: Real-time browser operation preview.
View past tasks with search functionality and playback capabilities.
Comprehensive settings interface with 8 panels:
- General: Language, startup behavior, window preferences
- Providers: AI provider selection and API key management
- Chat: Temperature, max tokens, search provider configuration
- Agent: Browser/File/Custom agent settings, per-agent MCP services
- Scheduled Tasks: Create and manage automated tasks
- User Interface: Theme, font size, density preferences
- Network: Proxy, timeout, and retry configuration
- Memory: Context management settings (coming soon)
- DeepSeek: deepseek-chat, deepseek-reasoner
- Qwen (Alibaba Cloud): qwen-max, qwen-plus, qwen-vl-max
- Google Gemini: gemini-2.0-flash, gemini-2.5-pro, gemini-2.5-flash, and more
- Anthropic Claude: claude-sonnet-4-5, claude-3-7-sonnet, claude-3-5-sonnet, and more
- OpenRouter: Multiple providers (Claude, GPT, Gemini, Mistral, Cohere, etc.)
- Configuration Guide - Detailed API key setup instructions
- Testing Guide - E2E testing with Playwright
Special thanks to Eko - A production-ready agent framework that powers the AI capabilities of this project.
⭐ If you find this project helpful, please consider giving it a star! Your support helps us grow and improve.
- Report issues on GitHub Issues
- Join discussions and share feedback
- Contribute to make AI Browser better
Contributions are welcome! Please read our contributing guidelines before submitting pull requests.
Important: Never commit actual API keys to the repository. Use the Settings UI to configure your development environment.
This project is licensed under the MIT License - see the LICENSE file for details.