🍑 Peach

A high-performance, private, and zero-cost RAG (Retrieval-Augmented Generation) engine that runs entirely in your browser.

🚀 Key Features

• Zero Cost: No server-side GPUs, no API keys, and no subscription fees. Everything runs on the client machine.
• Orma Indexing: Ultra-fast, browser-native vector search powered by Orama. Handles thousands of documents with sub-millisecond search times.
• WebLLM: First-class support for on-device LLMs via WebLLM and WebGPU. Stream answers directly from local models like Qwen2.5.

🛠️ Stack

• Framework: Angular 19
• Vector Search: Orama (Sharded & lazy-loaded)
• Embeddings: Transformers.js (all-MiniLM-L6-v2)
• Inference: WebLLM (Qwen2.5-0.5B-Instruct)

🏁 Quick Start

1. Install Dependencies

   npm install
   cd indexer && npm install && cd ..

2. Build Search Index Place your .md files in content/, then run:

   cd indexer && node build-index.mjs

3. Run Locally

   npm start

   Open http://localhost:4200 in your browser. Note: A WebGPU-enabled browser (like Chrome 113+) is required for on-device LLM inference.

🚀 Deployment

To deploy to GitHub Pages:

npm run deploy

This will build the project with the correct base-href (/Peach/) and push the dist/slm/browser folder to the gh-pages branch.

🔒 Privacy First

• Local Inference: Your questions and the LLM's answers are processed entirely on your machine via WebGPU.
• No Tracking: No user data, queries, or chat history are ever sent to a server.

---

License: MIT