Touch Grass — A Bird ID App That Works When There's No Signal

Description and use

Touch Grass is a PWA that uses open‑weight AI models to identify birds, plants, and insects entirely on-device, enabling offline, private identification.

Problem

Identifying bird calls typically requires uploading audio to a cloud API, exposing location and usage data.

Outcome

The app enables quick, private bird identification offline, preserving user privacy and allowing use in low‑signal environments.

Workflow

• Record a 3‑second audio clip
• Convert clip to a mel spectrogram locally
• Run BirdNET‑ONNX model in a Web Worker
• Return species prediction without any network call

Evidence

Touch Grass is a Progressive Web App that runs open‑weight AI models entirely inside the browser.

Source

Record 3-second audio clips, identify species via BirdNET‑ONNX (~20MB, quantized, 3000+ species), all on‑device.

Source

After initial model downloads, no internet required. Service worker caches the app shell; IndexedDB stores observations and model cache.

Source