Skip to main content
Back

CJ Pais Built Handy After Breaking a Finger, Now It Has 21K Stars

Handy is a free, MIT-licensed, fully offline speech-to-text desktop app built in Rust that pastes transcriptions directly into any text field — no cloud, no account, no cost.

  • Free and fully offline: Handy is an MIT-licensed, open-source speech-to-text desktop app with no cloud, no account, and no cost.
  • Two model backends: Supports OpenAI Whisper (Small/Medium/Turbo/Large) with GPU acceleration, and NVIDIA Parakeet V3 for CPU-only, 25-language transcription.
  • Parakeet V3 is the standout: A 600M-parameter model running at up to 5x real-time speed on mid-range CPUs, with automatic language detection across 25 European languages.
  • Built lean with Tauri: Rust backend + React frontend; the Windows installer is just 12.6 MB — far lighter than Electron alternatives.
  • Linux's best dictation option: Nearly no polished dictation app covers Linux; Handy has been widely adopted there, earning 200+ points on Hacker News.
  • Known limits: No real-time streaming (2–5s latency after speaking), occasional Whisper crashes on Windows/Linux, and Wayland requires extra setup tools.

Handy started because its creator, CJ Pais, broke a finger. Stuck in a cast and unable to type properly, he tried the usual suspects , MacWhisper, SuperWhisper , and found them either paywalled or opaque about where audio was going. So he built Handy: a free and open source speech-to-text application that runs on Windows, macOS, and Linux. It has since crossed 21,000 GitHub stars and is still actively shipping.

One shortcut, any text field

Handy lets users start transcription instantly using a configurable keyboard shortcut , press to record, release to transcribe , and automatically pastes the resulting text into any active text field. There is no dedicated editor, no clipboard manager to configure, and no cloud account to authenticate. There is nothing to buy. Handy has no paid tier, no subscription, no word limits, and no account system.

The workflow is intentionally minimal:

  1. Press your configured hotkey (push-to-talk or toggle mode)
  2. Speak
  3. Release , Handy processes locally and pastes the result into whatever app has focus

Two model families, one offline pipeline

Handy's backend leverages OpenAI's Whisper models for GPU-accelerated speech recognition and Parakeet V3 for efficient CPU-only transcription with automatic language detection. Before any audio even reaches the model, Handy integrates Silero's Voice Activity Detection (VAD) for silence filtering, ensuring only speech segments are processed. VAD (Voice Activity Detection) is a lightweight classifier that strips out silence before it hits the heavier transcription model , this keeps latency low and prevents Whisper from hallucinating text during quiet gaps.

The model options available in Handy today:

  • Whisper Small (487 MB) , fast, modest accuracy, runs on most hardware