EchoNote 1.0.1 — Gemma 4

EchoNote 1.0.1 — now powered by Google's Gemma 4.

• Gemma 4 E2B Instruct (Apache-2.0) is the new default model for organizing notes, running fully locally through llama.cpp. It classified all 8 test sentences correctly.
• Faster structured output: a compact grammar plus disabling "thinking" cut Gemma's time per note from ~14 s to ~8.6 s on the test laptop (CPU only).
• Qwen2.5 1.5B stays available in Settings as the faster option (~5.5 s per note); Gemma 3 1B and Qwen2.5 0.5B are available for low-memory machines.

Install (Linux x86-64, tested on Fedora 44 / GNOME 50)

chmod +x EchoNote-1.0.1-x86_64.AppImage
./EchoNote-1.0.1-x86_64.AppImage

The first launch downloads and SHA-256-verifies the models (2.7 GB) from Hugging Face; after that EchoNote works offline. Requirements: x86-64 CPU with AVX2, ~4.5 GB free RAM for the default models, ~3 GB disk.

Upgrading from 1.0.0: your notes are kept. EchoNote switches to Gemma 4 E2B, which needs a one-time 2.65 GB download in Settings → AI model; until it's installed, new notes are transcribed and wait in the Inbox. To stay on Qwen2.5 1.5B instead, select it in Settings → AI model.