Handy — Free Offline Speech-to-Text That Runs Anywhere

Introduction

Handy is a free, open-source speech-to-text application that processes all audio locally on your device. Built with Rust and Tauri, it delivers fast transcription without sending any data to external servers, making it suitable for sensitive environments and air-gapped workflows.

What Handy Does

• Transcribes speech to text in real-time using local models
• Runs entirely offline with zero cloud dependencies
• Supports multiple languages via Whisper-based models
• Provides a minimal desktop UI across macOS, Windows, and Linux
• Offers accessibility features for users who need speech input

Architecture Overview

Handy uses a Tauri v2 shell wrapping a Rust backend that runs Whisper-derived models for inference. Audio capture happens through platform-native APIs, and the transcription pipeline runs on CPU or GPU depending on available hardware. The lightweight frontend is built with standard web technologies rendered in the native webview.

Self-Hosting & Configuration

• Download prebuilt binaries from GitHub Releases (no build required)
• Models are bundled or downloaded on first launch (one-time)
• Configure language, model size, and audio input device in settings
• No accounts, API keys, or internet connection needed after setup
• Build from source with Rust toolchain and Tauri CLI

Key Features

• Complete offline operation preserving privacy
• Cross-platform support via Tauri (macOS, Windows, Linux)
• Low latency real-time transcription
• Extensible architecture for adding new model backends
• Small application footprint compared to Electron alternatives

Comparison with Similar Tools

• Whisper (OpenAI) — Python library requiring manual integration; Handy provides a ready-to-use desktop app
• whisper.cpp — C++ inference only; no GUI or audio capture built in
• macOS Dictation — Platform-locked and sends data to Apple servers
• Otter.ai — Cloud-based with subscription; Handy is free and offline
• Faster Whisper — Server-focused; Handy targets end-user desktop use

FAQ

Q: What models does Handy use internally? A: It uses Whisper-derived models optimized for local inference, with options for tiny through large model sizes.

Q: Does it work on Apple Silicon? A: Yes. Handy supports macOS on both Intel and Apple Silicon natively.

Q: Can I use it for long-form transcription? A: Yes. It handles continuous recording sessions and exports transcripts as plain text.

Q: Is GPU acceleration supported? A: On supported hardware, Handy uses GPU acceleration for faster inference. CPU-only mode works on all machines.

Sources

• https://github.com/cjpais/Handy