Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

Handy Review: Free Offline Speech-to-Text Dictation App

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Handy (version 0.9.8) is a free, MIT-licensed desktop app that turns your speech into text on your own computer and pastes it into the text field you are typing in. It runs on Windows, macOS and Linux, needs a one-time model download, and every statement here comes from the maker's website, privacy policy and GitHub repository.

Key Takeaways

  • Maintenance: the latest release was published on 3 October 2026 and the main branch last changed on 21 October 2026, so no stale-project warning applies.
  • Cost: no paid tier; the project asks for donations and lists sponsors.
  • Models: Whisper (Small, Medium, Turbo, Large) and Parakeet, downloaded once and run locally.
  • Installers: .dmg, .exe, .msi, .AppImage, .deb and .rpm files on the GitHub Releases page.
  • Maker: CJ Pais with contributors; the project calls itself forkable rather than the best dictation app.

📍 In One Sentence

Handy is an open-source desktop app that records your voice when you press a shortcut, transcribes it with a local speech model and pastes the text into the active app.

💬 In Plain Terms

You press a key combination, say a sentence, let go, and the words appear where your cursor is; according to the project, no account and no internet connection are needed once the model is on your disk.

📌Note: This review is based on handy.computer, its privacy policy and the GitHub repository, checked on 24 October 2026. PromptQuorum has not installed or timed the app, so accuracy and speed are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Handy is a free, open-source desktop app by CJ Pais and contributors: you hold a shortcut, speak, release, and the transcribed text is pasted into the text field you are using. The project says speech recognition runs on your own computer and works completely offline once a model is downloaded. This review is based on the website, its privacy policy and the GitHub repository, checked on 24 October 2026; PromptQuorum has not tested the app hands-on.

Handy Review: Free Offline Speech-to-Text Dictation App

What Is Handy?

Handy is a cross-platform desktop speech-to-text app that records while you hold a configurable shortcut, transcribes the audio locally and pastes the result into whatever application has focus. The project describes itself as free, open source, private and simple: "one tool, one job". The code is MIT-licensed and the repository was created in February 2025.

The About page says Handy is not trying to be the best speech-to-text app but "the most forkable one", and invites contributors who know Rust. It is a downloadable program, not a hosted service.

Get It

Handy is a desktop download for macOS, Windows and Linux; no iOS, Android or web version is listed.

Option
How to get it
macOS.dmg (Apple silicon, Intel): handy.computer
Windows.exe, .msi (x64, ARM64): handy.computer
Linux.AppImage, .deb, .rpm (x86_64, ARM64): handy.computer
Websitehandy.computer
Privacy policyhandy.computer/privacy
Source codegithub.com/cjpais/Handy, MIT license

Version reviewed: 0.9.8, published on 3 October 2026 per the GitHub Releases page, which also lists every installer. The website's main button offers the Apple silicon build for Mac. Homebrew (brew install --cask handy) and winget (winget install cjpais.Handy) packages exist but are not maintained by the developers; on Debian or Ubuntu, install the .deb with sudo apt install ./Handy_*.deb. The app is free.

Requirements

The website lists no minimum system versions; hardware needs depend on which speech model you pick.

Operating system

What the sources say:
Website silent on minimums; the project's build configuration sets macOS 10.15 as the minimum; downloads exist for Windows x64 and ARM64 and Linux x86_64 and ARM64

Speech models

What the sources say:
Downloaded inside the app: Whisper Small 487 MB, Medium 492 MB, Turbo 1600 MB, Large 1100 MB, and a 731 MB Parakeet model, per the project's docs

Hardware

What the sources say:
Whisper models use GPU acceleration when available; Parakeet is described as CPU-optimized; no RAM figure is given

Permissions

What the sources say:
Microphone and accessibility access, requested at first launch

Linux extras

What the sources say:
xdotool (X11), wtype or dotool (Wayland) for typing text; libgtk-layer-shell at runtime

Internet

What the sources say:
Needed to download models and, if enabled, to check for updates; model files can also be copied in by hand

Model sizes are the file sizes in the project's manual-install guide; the in-app list may differ.

How to Get Started

The steps follow the project's README; PromptQuorum has not run them.

  1. 1
    Install Handy
    Why it matters: Download the installer for your system from the website or the Releases page and install it.
  2. 2
    Grant permissions
    Why it matters: Launch Handy and allow microphone and accessibility access when the system asks.
  3. 3
    Download a model
    Why it matters: Open Settings, then Models, and download a Whisper or Parakeet model; this is the main step that needs the internet.
  4. 4
    Dictate
    Why it matters: Click into any text field, hold the shortcut, speak and release; Handy pastes the transcript. Push-to-talk can be turned off so the same shortcut toggles recording instead.

Features Confirmed by the Sources

Every item below comes from the website, the privacy policy or the README; none has been independently tested.

  • Push-to-talk or toggle. The shortcut is configurable; hold-to-record is the default and a tap-to-start, tap-to-stop mode is available.
  • Local models. Whisper in four sizes and Parakeet V3, which the README describes as CPU-optimized with automatic language detection; custom Whisper GGML files placed in the models folder appear under Custom Models.
  • Voice-activity filtering. Silero VAD removes silence before transcription.
  • Custom words and history. Settings include a custom-word list, custom prompts and a transcription history that keeps recordings.
  • Optional post-processing. A second pass can clean up the transcript with an AI model you configure, toggled with handy --toggle-post-process; it is off unless you set it up.

Data Flow and Privacy

The privacy policy, effective 18 September 2026, says Handy needs no account, contains no advertising, behavioral analytics or tracking telemetry, and does not send your microphone audio or transcripts to the maintainers during ordinary transcription. Recordings, transcripts, history, settings, custom words and logs stay on your device; by default Handy keeps a limited number of recent entries with their audio, and you can change the retention period or set the history limit to zero. Official release builds redact transcript text from logs.

The app does use the network in three cases: update checks against GitHub Releases (on by default, can be switched off), model downloads from Hugging Face or, as a fallback, blob.handy.computer via Cloudflare, and optional post-processing. If you configure a cloud provider for post-processing, the transcript, prompt and your credential go directly from your computer to that provider, which the policy says may be OpenAI, Anthropic, OpenRouter, Groq, Cerebras, Z.AI or AWS Bedrock; a local or on-device endpoint avoids that.

📌Note: These are the maintainers' declarations, not audit results. The code is public under the MIT license, but PromptQuorum has not read it or inspected the app's network traffic.

Who Should Use It

  • Writers and developers who want to dictate into editors, chat boxes and terminals without a cloud service or subscription.
  • Accessibility users and tinkerers who want a small, open codebase they can fork or script.
  • Privacy-minded users who accept a model download and want transcripts kept on their own machine.

What We Could Not Verify

  • Accuracy and speed. PromptQuorum did not measure word error rates, latency or language coverage; the number of supported languages is not given in the sources we read.
  • Hardware performance. Run times on CPU-only machines, older Macs and different GPUs were not tested, and no RAM requirement is published.
  • Platform caveats the project lists itself. The README warns that a Bluetooth headset microphone on macOS can lower playback quality, that fn and Globe shortcuts work only on Apple keyboards, and that Wayland support on Linux is limited.
  • Installer signing. Whether the macOS and Windows installers are notarized or signed with a developer certificate was not checked.
  • Pre-1.0 status. The version is 0.9.8, and the README says the project is actively developed with known issues.

Competitors and Alternatives

MacWhisper

Platforms:
macOS
Price / license:
Freemium / Proprietary
Key difference:
Mac-only app with a one-time purchase for local Whisper and Parakeet transcription; closed source
Articles about MacWhisper (2)

Also mentioned in:

TranscriptionSuite

Platforms:
macOS, Windows, Linux
Price / license:
Free / GPL-3.0
Key difference:
Adds live transcription, speaker diarization and a searchable audio notebook; the engine runs in Docker on Windows and Linux
Articles about TranscriptionSuite (1)

AquaVoice

Platforms:
macOS, Windows, iOS, Android
Price / license:
Freemium / Proprietary
Key difference:
Cloud dictation service that types into the active app, so it is not an offline option
Articles about AquaVoice (2)

These are same-category apps from the PromptQuorum directory; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

Can Handy run on a computer that is offline?

Yes, once models are in place: the project documents copying model files into the app's models folder by hand, and Handy also picks up models already in the shared Hugging Face cache. Custom Whisper GGML .bin files placed there appear under Custom Models.

Which model should I pick?

The README describes Whisper models as GPU-accelerated and Parakeet as CPU-optimized with automatic language detection, and the manual-install guide marks a Parakeet Unified EN 0.6B file as recommended; the sources give no benchmark between them.

Does Handy work on Wayland?

Support is limited: the README says typing needs wtype or dotool, global shortcuts must be set in your desktop environment to run handy --toggle-transcription, and Ubuntu 26.04 needs ydotool instead of wtype.

Can I control Handy from scripts or Raycast?

Yes: flags such as --toggle-transcription, --toggle-post-process and --cancel control a running instance, and the README links a Raycast extension by a third-party developer that starts and stops recording.

Are Handy's releases signed?

The README says release artifacts use Tauri's updater signature format with the public key stored in the repository's tauri.conf.json, and the Releases page lists a .sig file next to every installer.

Verdict

Handy is a small, actively updated, free and open-source way to dictate into any app on Windows, macOS or Linux, with speech recognition that by the project's account runs on your own computer. Against that, it is still below version 1.0, the project itself lists macOS Bluetooth and Linux Wayland caveats, and none of its accuracy or speed could be checked here. It suits people who want push-to-talk dictation without an account or a subscription and can try it on their own hardware; anyone who needs proven accuracy, a polished Wayland experience or vendor support should test it first.

Sources

  • Handy website — description, download button and sponsors, checked 24 October 2026.
  • Handy About page — project goals and how it works, checked 24 October 2026.
  • Handy privacy policy — data handling, network use and post-processing providers, effective 18 September 2026.
  • Handy on GitHub — README, MIT license, installation notes and known issues, checked 24 October 2026.
  • Handy releases — version 0.9.8 and installer files, checked 24 October 2026.

← Back to Power Local LLM

MCP