Touch Grass AI: Offline Outdoor Companion & Open-Source Species Identifier

Description and use

Touch Grass AI is an offline‑first web app that identifies plants and birds, plans routes, and logs observations using ViT, MobileNetV3, AST, Leaflet, and OpenStreetMap.

Problem

Users want to identify plants and birds while hiking or gardening without relying on cellular data or paid cloud APIs, and they also want to reduce screen time.

Outcome

The application enables offline plant and bird identification, route planning, and species journaling, reducing reliance on cloud services and encouraging outdoor engagement.

Workflow

• Capture plant image with the web client
• Send image to /api/classify-image for ViT/MobileNetV3 inference
• Receive species prediction and display result
• Record audio with the web client
• Send audio to /api/classify-audio for AST inference
• Receive bird‑call prediction and display result
• Use Leaflet to build or view routes
• Export routes as GPX
• Log observations in local offline journal

Evidence

Touch Grass AI is an open-source, offline-first outdoor companion designed to help people reconnect with nature while reducing unnecessary screen time.

Source

The application is organized into two main parts: 1. Client-side experience 2. Backend inference and APIs

Source

Vision Transformer (ViT) and MobileNetV3 are the proposed vision models for classifying plant-related images

Source

The Audio Spectrogram Transformer (AST) is used in the planned audio-classification pipeline for analyzing recorded sounds and supporting bird-call recognition.

Source

The expected response is: {"status":"online","appName":"Touch Grass AI","openAIWeights":"Pre-Warmed Ultra-Fast ViT & AST Neural Engine (<15ms)","offlineModeSupported":true,"timestamp":"2026-10-10T08:00:00.000Z"}

Source