gemma-gem — What is it?

Gemma Gem is a browser-based AI assistant that runs Google's Gemma 4 model on-device using WebGPU, ensuring privacy and eliminating the need for cloud services.

⭐ 439 Stars 🍴 33 Forks TypeScript Apache-2.0 Author: kessler
Source: per README View on GitHub →

Why it matters

Gemma Gem is gaining attention due to its focus on privacy and on-device AI processing, which addresses the pain points of data privacy concerns and the limitations of cloud-based AI services. Its unique technical choice of using WebGPU for on-device inference stands out, as it allows for real-time AI interactions without data leaving the user's machine.

Source: Synthesis of README and project traits

Core Features

On-device AI processing

Gemma Gem leverages WebGPU to run the Gemma 4 model on the user's device, ensuring that no data leaves the machine and providing a private AI experience.

Source: per README
Browser extension

Gemma Gem is a browser extension that integrates with Chrome, allowing users to interact with the AI assistant directly within their browser.

Source: per README
Model selection

Users can choose between the E2B and E4B versions of the Gemma 4 model, depending on their hardware capabilities and desired performance.

Source: per README

Architecture

The architecture of Gemma Gem is modular, with distinct components for offscreen document processing, service worker message routing, and content script interaction with the DOM. It employs a message-passing system between these components, utilizing WebGPU for model inference and service workers for background tasks like screenshot capture and JavaScript execution.

Source: Code tree + dependency files

Project Knowledge Graph

Knowledge graph: project (center) + core features (inner hexagons) + key dependencies (outer chips) @huggingface/transformers@huggingface/t… @kessler/gemma-agent@kessler/gemma… marked On-device AI processingOn-device AI proces… Browser extension Model selection gemma-gem Project Core feature Key dependency

Center: project; inner ring: core feature modules; outer ring: key dependencies. Auto-generated from core_features and tech_stack.key_deps.

Tech Stack

LanguageTypeScriptFrameworkWXT (Chrome extension framework), @huggingface/transformers (browser ML inference)
@huggingface/transformers@kessler/gemma-agentmarked
Chrome browser extension
Source: Dependency files + code tree

Quick Start

```bash pnpm install pnpm build Load the extension in `chrome://extensions` (developer mode) from `.output/chrome-mv3-dev/`. ```
Source: README Installation/Quick Start

Use Cases

Gemma Gem is suitable for developers and users who require an AI assistant for web-based tasks, such as reading web pages, interacting with web applications, or automating web interactions. It is particularly useful for scenarios where privacy and on-device processing are critical, such as in corporate environments or for personal use on sensitive websites.

Source: README

Strengths & Limitations

Strengths

  • Strength 1: Strong focus on privacy and on-device processing
  • Strength 2: Easy to integrate into existing workflows as a browser extension
  • Strength 3: Offers flexibility with model selection

Limitations

  • Limitation 1: Hardware requirements may be high for optimal performance
  • Limitation 2: Limited to Chrome browser
  • Limitation 3: May have performance limitations compared to cloud-based solutions
Source: Synthesis of README, code structure and dependencies

Latest Release

0.3.0 (no release date provided), Main changes: Initial release with core features.

Source: GitHub Releases

Verdict

Gemma Gem is a promising project for those seeking a private and efficient AI assistant for web-based tasks. Its focus on on-device processing and privacy makes it a standout choice for users concerned about data security. It is particularly well-suited for developers and users who require AI capabilities within the browser without relying on cloud services.

Transparency Notice
This page is auto-generated by AI (a large language model) from the following public materials: GitHub README, code tree, dependency files and release notes. Analyzed at: 2026-05-23 18:31. Quality score: 85/100.

Data sources: README, GitHub API, dependency files