Home

Zerm

Zerm is a native macOS app that turns your voice into clean text and reads text back to you — built around three on-device AI models that the app downloads and manages for you, so the core experience works offline and private.

Local-first, ludicrously fast voice-to-coding-agent app for developers.

Zerm is a GPLv3 derivative of VoiceInk by Beingpax, adapted for Arcusis branding and product direction. See NOTICE.

The three on-device models

Model	Job	Engine
🎙️ Whisper	Speech-to-Text (dictation)	whisper.cpp (ggml)
🔊 Kokoro	Text-to-Speech (Read Aloud)	sherpa-onnx (Kokoro-82M)
🧠 Gemma	Agentic layer (smart reading + enhancement)	llama.cpp (Gemma 4 E2B)

Each model installs the same way: pick it, the app downloads it once with a progress bar, then it runs fully offline. Cloud providers remain available for every task as an option. See The Three-Model Platform and Model Management.

What Zerm does

• Dictation — press a hotkey, speak, and Zerm transcribes and pastes at your cursor. See Speech-to-Text.
• Read Aloud — select text anywhere, press a hotkey, and Zerm reads it in a natural voice. See Read Aloud.
• Smart Reading — text is cleaned (acronyms, URLs, code, emoji, tables) and optionally rewritten by the on-device LLM so it sounds human, not robotic. See Smart Reading.
• AI Enhancement — dictated text can be cleaned/reformatted by an LLM (on-device or cloud). See AI Enhancement.
• Power Mode — app- and URL-aware prompt selection.

Start here

• Architecture — the big picture and how the pieces fit
• Build & Release — building from source, the Makefile, releases
• Privacy Model — what stays on-device and what doesn't

Platform

macOS 14.4+ (Apple Silicon recommended). Native Swift/SwiftUI + AppKit. No Windows/Linux build.