What it is
Handy is a desktop dictation app by one developer, CJ Pais (cjpais on GitHub), first published in February 2025. In its own words: “A free, open source, and extensible speech-to-text application that works completely offline.”
- Push to talk: hold your shortcut, speak, release. The text is pasted into the app you’re in: email, chat, a doc, a code editor.
- On your computer: the speech model is a file on your disk. Your audio isn’t sent to a server to be turned into text.
- Your choice of model: Parakeet V3 (recommended, 456 MB, 25 European languages), Whisper Small to Large, Moonshine and more.
- Free: no account, no word limit, MIT license.
| Handy | Wispr Flow | |
|---|---|---|
| Price | $0 | Pro $15/month ($12 billed yearly); free plan 2,000 words/week on desktop |
| Where speech becomes text | on your computer | Flow’s servers (its security page: audio “streams … to our servers”) |
| Cleans up “um” and formats | optional, you set it up (section 05) | yes, automatically |
| Phone app | no (desktop only) | yes |
Before you start
| Mac | Apple Silicon or Intel builds. Needs the Microphone and Accessibility permissions (Accessibility is how it pastes the text). |
| Windows | x64 and ARM64 installers (.exe or .msi). |
| Linux | AppImage, .deb, .rpm. On Wayland you also need wtype or dotool (the README explains). |
| Disk | About half a gigabyte for the recommended model; Whisper Large is bigger. |
| Bluetooth headsets | On Mac, a Bluetooth headset mic can drop to low quality while recording; the README lists it as a known issue. The built-in mic works fine. |
Install it safely
Get it only from the developer: handy.computer or the releases page, github.com/cjpais/Handy/releases. GitHub shows a SHA-256 checksum next to each file; checking it takes one minute.
Mac
- Download
Handy_<version>_aarch64.dmg(Apple Silicon) orHandy_<version>_x64.dmg(Intel). - In Terminal, run the command below and compare the result with the sha256 shown next to that file on the releases page. They must match exactly. (v0.9.8 Apple Silicon:
81ac05aa…20779417.)
shasum -a 256 ~/Downloads/Handy_*_aarch64.dmg
- Drag Handy into Applications and open it. If macOS says it can’t check the app, go to System Settings › Privacy & Security and click Open Anyway (only for a file whose checksum matched).
- Allow Microphone and Accessibility when it asks.
Homebrew works too: brew install --cask handy. The developers note that the cask isn’t maintained by them.
Windows
- Download
Handy_<version>_x64-setup.exe(or the arm64 one). - In PowerShell, in your Downloads folder, run the command below and compare it with the sha256 on the releases page.
Get-FileHash .\Handy_*_x64-setup.exe -Algorithm SHA256
Pick the model, set the key
- On first launch Handy downloads a speech model once. Pick Parakeet V3: it’s the one Handy recommends (fast on a normal CPU, 25 European languages, picks the language for you).
- English only and want it lighter? Moonshine Base is very fast. Want the most accuracy and have a good GPU or Apple Silicon? Try a Whisper Turbo or Large model.
- Set your shortcut in Settings. A combination like
option+spaceworks on any keyboard (thefnkey only works on Apple keyboards). - Hold the shortcut, talk, let go. That’s it.
The two times it goes online
My AI agent searched Handy’s source (66 Rust files, 32,115 lines on Oct 10, 2026) for every web address and read those paths. It found two:
- Model downloads from
blob.handy.computer, once per model you pick. - The update check, a small file on GitHub (
…/releases/latest/download/latest.json). It’s on by default.
To switch the update check off: press ⌘⇧D (Windows and Linux: Ctrl+Shift+D) to open the debug settings, and turn off Check for Updates. Then check for new versions yourself now and then on the releases page.
The optional AI cleanup (next section) is off by default. If you connect it to a cloud provider, your text goes to that provider; on a Mac with Apple Intelligence it can run on your own machine.
This was a code read, not a full security audit.
Clean up the ums (optional)
Out of the box Handy types exactly what you said. To get Wispr-style clean text, turn on Post Processing (in the same debug settings), pick a provider (Apple Intelligence runs on-device on Apple Silicon Macs with macOS 26 or later; OpenAI, Anthropic, Groq, OpenRouter and others need an API key), and create a prompt. Handy puts your transcript where you write ${output}.
Clean up this dictated text. Remove filler words (um, uh, like, you know, so at the start), false starts and repeated words. Fix punctuation and capitalization. Keep my words, my meaning and my tone; don't add anything, don't summarize, don't make it formal. If I said a list, format it as a list. Return only the cleaned text.
${output}Checklist
- I downloaded from handy.computer or github.com/cjpais/Handy, nowhere else.
- The SHA-256 of my file matches the one on the releases page.
- I picked Parakeet V3 (or a model that fits my language).
- My shortcut uses standard keys (not
fnon an external keyboard). - I decided about the update check, on or off.
- If I use AI cleanup with a cloud provider, I know my text goes there.
A prompt for your AI agent
Paste this into Claude Code or any agent that can read files and use a terminal.
I downloaded Handy from the official releases page: https://github.com/cjpais/Handy/releases/latest. The file is at <PATH TO THE FILE>. Compute its SHA-256 and compare it with the sha256 GitHub lists for that file name on the release page (you can read it with: gh api repos/cjpais/Handy/releases/latest). Show me both values and say whether they match. Don't open, install or run anything.
Sources
- github.com/cjpais/Handy: README, stars, license, releases (GitHub API, Oct 11, 2026); source read Oct 10, 2026
- handy.computer: the official site
- wisprflow.ai/pricing and Wispr Flow’s privacy & security overview (Oct 10, 2026)
- Whistle, the 16.9 MB speech model from the reel (Cactus Compute, Apache-2.0). Handy doesn’t ship it.