Feelyai
AI in your browser that supports MCP Servers, no network call, no costs, 100% private
Installation
npx feelyaiAsk AI about Feelyai
Powered by Claude Β· Grounded in docs
I know everything about Feelyai. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
FeelyAI
Run advanced AI models directly in your browser. Private, secure, and purely local.
FeelyAI is a powerful, privacy-focused AI chat application that runs entirely in your browser using WebLLM. No server costs, no data sent to external servers, and 100% client-side execution.
π Try Live Demo
Experience FeelyAI in action at feelyai.com
β¨ Features
π 100% Private & Local
- All AI processing happens directly in your browser using WebGPU
- No data sent to external servers
- No API keys required
- Complete privacy and security
π€ Multiple AI Models
Choose from several optimized language models:
- Llama 3.2 3B - Fast and efficient general-purpose model
- Hermes 3 (Llama 3.1 8B) - Advanced reasoning and tool use
- TinyLlama 1.1B - Ultra-lightweight for quick responses
- Qwen2.5-Coder-3B - Specialized for coding tasks
π οΈ MCP (Model Context Protocol) Integration
- Connect to MCP servers to extend AI capabilities
- Add custom tools and functions
- Support for custom headers and authentication
- Enable/disable servers on the fly
- Real-time tool discovery and schema introspection
π§ Built-in Tools
FeelyAI includes powerful built-in tools:
evalCode- Execute JavaScript code directly in the browser (with safe eval mode)listTools- Discover available tools with search and regex filteringgetToolSchema- Inspect tool schemas before use
π― Smart Tool Execution
- Schema Discovery Pattern - AI automatically discovers and learns tool schemas
- Validation & Error Handling - Automatic validation with helpful error messages
- Permission System - Granular control over tool execution
- Allow once
- Allow for session
- Always allow
- Safe Eval Mode - Run code in isolated Web Workers for security
π¬ Advanced Chat Management
- Projects & Chats - Organize conversations into projects
- π Encrypted Projects - Password-protect sensitive projects with client-side AES-GCM encryption
- All chats, messages, and system prompts encrypted locally
- Auto-lock on page reload for security
- Passwords hashed with bcryptjs (never stored in plain text)
- Unique encryption keys per project using PBKDF2
- Persistent Storage - All chats saved locally using IndexedDB
- Auto-titling - Chats automatically named from first message
- Message History - Full conversation history with tool calls and outputs
- Markdown Support - Rich text rendering with syntax highlighting
βοΈ Customizable System Prompts
- Per-Project System Prompts - Each project can have its own custom system instructions
- Edit system instructions to customize AI behavior
- Template variables for dynamic tool injection (
{{listTools}},{{tool_names}}) - Reset to defaults anytime
- Persistent across sessions
- Encrypted along with project data for password-protected projects
π¨ Modern UI/UX
- Clean, dark-themed interface
- Real-time progress indicators during model loading
- Collapsible tool outputs for better readability
- Responsive design
- Smooth animations and transitions
π Model Switching
- Switch between models on the fly
- Automatic model download and caching
- Progress tracking during model initialization
π Getting Started
Prerequisites
- Modern browser with WebGPU support (Chrome 113+, Edge 113+)
- Sufficient RAM (4GB+ recommended)
- GPU with WebGPU support
Installation
- Clone the repository:
git clone https://github.com/hasmcp/feelyai.git
cd feelyai
- Install dependencies:
npm install
- Run the development server:
npm run dev
- Open your browser and navigate to
http://localhost:5173
Building for Production
npm run build
npm run preview
π Adding MCP Servers
- Click the "Add Server" button in the sidebar
- Enter the MCP server URL (e.g.,
http://localhost:3000/mcp) - Optionally add a name and custom headers
- Click "Add Server"
The AI will automatically discover available tools from connected servers.
π‘οΈ Security Features
Safe Eval Mode
When enabled (recommended), JavaScript code execution happens in an isolated Web Worker, preventing access to:
- DOM manipulation
- Local storage
- Cookies
- Parent window context
Permission System
Control which tools the AI can execute:
- Allow Once - Single execution approval
- Allow for Session - Approve for current session
- Always Allow - Trust tool permanently
π¦ Tech Stack
- Frontend Framework: Vue 3 (Composition API)
- Build Tool: Vite
- AI Engine: @mlc-ai/web-llm
- MCP Client: @modelcontextprotocol/sdk
- Styling: TailwindCSS 4
- Icons: Lucide Vue
- Markdown: markdown-it
- Storage: IndexedDB (via idb)
- Validation: Ajv (JSON Schema)
- Encryption: bcryptjs + Web Crypto API (AES-GCM)
ποΈ Architecture
src/
βββ components/
β βββ AddServerModal.vue # MCP server configuration
β βββ ProjectSidebar.vue # Project/chat navigation
β βββ McpIcon.vue # MCP branding
βββ composables/
β βββ useChat.js # Main chat logic & state
βββ services/
β βββ McpClient.js # MCP protocol client
β βββ MessageStore.js # IndexedDB persistence
β βββ EncryptionService.js # Client-side encryption
βββ workers/
β βββ llm.worker.js # WebLLM worker
β βββ eval.worker.js # Safe code execution
βββ App.vue # Main application
π€ Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
Created by
HasMCP - No-code/low-code MCP server framework with built-in auth, realtime logs and telemetry
π License
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.
π Acknowledgments
- WebLLM - Browser-based LLM inference
- Model Context Protocol - Tool integration standard
- MLC AI - Machine Learning Compilation
π Links
β οΈ Disclaimer
Usage of LLM models is subject to their respective license agreements. Please review the license terms for each model you use. FeelyAI is provided as-is, and AI-generated content may contain errors or inaccuracies.
Powered by WebLLM β’ No Server Costs β’ 100% Client-Side
