Skip to content

msbragi/eye-assistant

Repository files navigation

GemmaLink: Eye Assistant

GemmaLink is a local-first AI tool that turns your smartphone into a vision sensor for Gemma 4. It enables a "Zero-Footprint" multimodal experience, allowing you to discuss what your phone sees with a high-performance AI running entirely on your PC.

Designed for privacy and performance, it bypasses the cloud entirely using a high-concurrency Go backend.

Key Features

  • 100% Local: No cloud, no tunnels, no external dependencies.
  • Smart Discovery: Automatically selects the best binary for your CPU/GPU (Vulkan/AVX2).
  • Multimodal: Real-time visual analysis using Gemma 4.
  • Architectural Simplicity: Single-binary distribution (Go + Sidecar pattern).

Plug-and-Play Experience

GemmaLink is engineered for users of all technical levels. Unlike most local AI setups, it requires zero configuration:

  • No Python/Environment Setup: A single self-contained executable handles everything.
  • Auto-Provisioning: The app automatically finds, downloads, and configures the best AI engine for your specific hardware.
  • Instant Mobile Link: Just scan the generated QR code. The only "technical" step is a one-time confirmation in your mobile browser to trust the local secure connection.

Project Architecture

For a deep dive into the system design, components, and data flow, see the Architecture Documentation.

Build Instructions

The project uses a structured build system located in the scripts directory.

Prerequisites

  • Go 1.21+
  • Wails CLI
  • Task or standard Bash environment

Building for Production

Pass the target OS and version tag to the build script:

# General syntax: ./scripts/build.sh [OS] [Version]

# Build for Windows
./scripts/build.sh win 1.0.0

# Build for Linux
./scripts/build.sh linux 1.0.0

# Build all supported platforms
./scripts/build.sh all 1.0.0

Development Build

To load assets directly from the filesystem for rapid iteration:

./scripts/build-dev.sh

Releases & Downloads

  • Latest Binary (Windows/Linux): [Link Placeholder]
  • Source Code (zip/tar.gz): [Link Placeholder]

Disclaimer

This project is part of the Gemma 4 Challenge. It is intended for local deployment to demonstrate the capabilities of multimodal models in a private network environment.

Footprint note

Design Philosophy

GemmaLink: Eye Assistant is not intended to compete with cloud-scale vision services like Google Lens. Instead, it serves a different paradigm: Sovereign Local Intelligence.

While cloud services offer global-scale knowledge, GemmaLink provides:

  • Absolute Privacy: Your camera feed never touches a server outside your home.
  • Latency Control: Fast, local-loop processing independent of your internet bandwidth.
  • Extensibility: A focused tool for personal automation and private context analysis.