Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
On-device AI agent Chrome extension powered by Transformers.js and Gemma 4
| Date | Stars |
|---|---|
| 2026-07-31 | 1135 |
| 2026-08-01 | 1134 |
| 2026-08-06 | 1134 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Transformers.js Gemma 4 Browser Assistant ## About this extension An on-device AI assistant that runs entirely in your browser using WebGPU and Transformers.js. This Chrome extension provides an intelligent agent that can understand natural language commands and interact with your browser through a set of specialized tools. All processing happens locally on your device. No data is sent to external servers, ensuring complete privacy. ### What can it do? The AI agent has access to several tools that enable it to help you control and navigate your browser: #### Tab Management The agent can manage your browser tabs through natural language: - **get_open_tabs**: List all open tabs with their titles, URLs, and descriptions - **go_to_tab**: Switch to a specific tab by ID - **open_url**: Open new URLs in background or foreground tabs - **close_tab**: Close specific tabs #### Website Interaction (RAG) The extension uses Retrieval-Augmented Generation to understand and interact with webpage content: - **ask_website**: Search and extract relevant information from the current webpage using semantic similarity. The content script extracts structured content (headings, paragraphs, lists), generates embeddings using all-MiniLM-L6-v2, and returns the most relevant sections based on your query. - **highlight_website_element**: Visually highlight specific elements on the page. The agent can direct your attention to specific content by highlighting and scrolling to relevant sections. #### History Vector Database The extension maintains a semantic search-enabled history database: - **find_history**: Search your browsing history using natural language queries instead of exact keywords. The system stores vector embeddings for page titles, descriptions, and URLs in IndexedDB, enabling semantic search with time-based filtering. ### Installation #### Prerequisites - Chrome browser with WebGPU support (Chrome 113+) - Modern GPU with WebGPU capabilities #### Setup 1. Clone the repository: ```bash git clone <repository-url> cd tfjs-agentgemma-extension ``` 2. Install dependencies: ```bash pnpm install ``` 3. Build the extension: ```bash pnpm run build ``` 4. Load in Chrome: - Open `chrome://extensions/` - Enable "Developer mode" - Click "Load unpacked" - Select the `dist` folder #### Development Mode For active development with automatic rebuilding: ```bash pnpm run dev ``` ### Usage 1. Click the extension icon to open the sidebar panel 2. On first use, the models will download automatically (one-time) 3. Once loaded, interact with the AI agent through the chat interface ### Permissions The extension requires these permissions: - `sidePanel`: Display chat interface - `activeTab`: Access current tab content - `storage`: Save settings and model cache - `scripting`: Inject content scripts - `tabs`: Needed to read the tab URL - `host_permissions`: Access webpage content on all URLs --- ## Gemma 4 This extension uses the `onnx-community/gemma-4-E2B-it-ONNX` instruction-tuned model from Hugging Face: - Model card: https://huggingface.co/onnx-community/gemma-4-E2B-it-ONNX - Format: ONNX (optimized for browser inference with Transformers.js + WebGPU) --- ## Extension Architecture This extension demonstrates an effective architecture for integrating Transformers.js into browser extensions. The design separates concerns across three key components, each optimized for specific tasks. ### Background Script: The AI Engine The background service worker hosts Transformers.js models as the centralized AI engine. **Why this works:** - **Persistent model loading**: Models are loaded once and shared across all tabs, side panels, and content scripts. This is crucial because loading multi-gigabyte models repeatedly would be impractical. - **Service worker lifetime**: Service workers can stay alive during active ML processing, which is essential for inference tasks that may take several seconds. - **Centralized processing**
Excerpt of 7,100 characters
Read on GitHub
Nico Martin · @huggingface · Switzerland
33
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:8e6270b18994ca14, desc:ai agent