| --- |
| title: AskMiku |
| emoji: 🌐 |
| colorFrom: blue |
| colorTo: pink |
| sdk: static |
| app_file: index.html |
| pinned: false |
| --- |
| |
| # Ask Miku |
|
|
| Ask Miku is a browser-based technical support chatbot featuring Retrieval-Augmented Generation (RAG) capabilities. It runs entirely client-side using WebAssembly, ensuring privacy and eliminating the need for a backend server. |
|
|
| ## Features |
|
|
| - **Fully Client-Side**: Powered by Hugging Face Transformers.js and WebAssembly. No data leaves your browser. |
| - **RAG Integration**: Uses a vector database to retrieve relevant technical context before generating responses, improving accuracy and reducing hallucinations. |
| - **Local Vector Storage**: Persistent storage using IndexedDB for fast retrieval of knowledge base entries. |
| - **Database Management**: Built-in web UI for importing, exporting, and managing the knowledge base. |
| - **WebGPU Acceleration**: Supports hardware acceleration for faster model inference. |
|
|
| ## Technology Stack |
|
|
| - **LLM**: gemma-4-E4B-it-ONNX (quantized q4f16) |
| - **Embeddings**: `Xenova/all-MiniLM-L6-v2` |
| - **ML Framework**: [Transformers.js](https://huggingface.co/docs/transformers.js) |
| - **Vector Database**: EntityDB |
| - **Storage**: IndexedDB |
|
|
| ## Getting Started |
|
|
| Since this is a static web application, you can run it using any local web server: |
|
|
| ```bash |
| # Example using static-web-server |
| static-web-server |
| ``` |
|
|
| 1. Open the application in a modern browser. |
| 2. Click **"Load model"** to initialize the LLM and RAG system. |
| 3. Start chatting with Miku! |
|
|
| ## Knowledge Base |
|
|
| The chatbot is powered by a specialized dataset of technical threads. |
|
|
| - **Dataset**: [lmg-neo-lora-v0.3](https://huggingface.co/datasets/quasar-of-mikus/lmg-neo-lora-v0.3) by quasar-of-mikus |
| - *Description*: An incomplete collection of threads from June 2023 - March 2025. |
|
|
| ## Credits |
|
|
| - **Vector Database**: This project uses [`entity-db.js`](https://github.com/babycommando/entity-db), an in-browser vector database wrapping IndexedDB and Transformers.js over WebAssembly. |
| - **Dataset**: Special thanks to quasar-of-mikus for the [lmg-neo-lora-v0.3](https://huggingface.co/datasets/quasar-of-mikus/lmg-neo-lora-v0.3) dataset. |
|
|
