Direct answer — Is Handy a good free dictation app in 2026? Yes, if you are comfortable with an open-source tool. Handy is free, MIT-licensed and runs Whisper or Parakeet V3 locally on Mac, Windows and Linux: press a shortcut, speak, and the text is pasted where you type. The trade-offs are community-only support, limited Wayland support on Linux, and optional AI cleanup that relies on cloud providers.
This Handy speech to text review looks at one of the most popular free dictation apps of 2026. Handy, available at handy.computer, has close to 34,000 stars on GitHub and regularly appears in lists of free Wispr Flow alternatives. It does one thing: turn your voice into text in any text field, without sending audio to the cloud.
Below, we check every claim against Handy’s own README, website and release notes, explain where its limits are, and compare it honestly with Weesper Neon Flow.
What is Handy?
The Handy dictation app is a desktop speech-to-text tool built by developer CJ Pais and a community of contributors. Its README describes it as “a free, open source, and extensible speech-to-text application that works completely offline.”
The workflow is deliberately minimal:
- Press a configurable shortcut (by default Option+Space on macOS, Ctrl+Space on Windows and Linux)
- Speak while holding the key (push-to-talk), or tap once to start and once to stop
- Release, and Handy pastes the transcription into the active app
Silence is filtered out with Silero VAD (voice activity detection, which detects when you are actually speaking). Handy’s website sums up the philosophy in three words: free, simple (“one tool, one job”) and private (“your voice stays on your computer”).
Under the hood, Handy is a Tauri app written in Rust and TypeScript. The latest release at the time of writing is v0.9.8, published on 3 October 2026.
Is Handy really free and open source?
Yes, on both counts. The code is published under the MIT licence, one of the most permissive open-source licences: you can read it, modify it, fork it and even reuse it in commercial software. Only the Handy name, logo and brand assets are excluded, so a fork must use its own branding.
There is no paid tier. Handy’s website states that “accessibility tooling belongs in everyone’s hands, not behind a paywall” and asks for optional donations and GitHub sponsorship instead.
That also means there is no company behind the product in the usual sense. Support happens through GitHub issues and the community, with no service-level commitment. For a free tool, that is fair; for a team that needs someone to call, it matters.
Does Handy work offline?
Yes, for transcription. Once you have downloaded a model, Handy transcribes speech entirely on your device, without an internet connection.
| Feature | What leaves your device |
|---|---|
| Transcription (Whisper, Parakeet) | Nothing: audio stays on your computer |
| Model download | A one-time download of the model file |
| AI post-processing (off by default) | Your transcript text, sent to the provider you choose |
The AI post-processing feature deserves a closer look. It is disabled by default. When you enable it, Handy can send your transcript to an LLM for cleanup through providers listed in its settings, including OpenAI, Anthropic, OpenRouter, Groq and Cerebras, Apple Intelligence on supported Macs, or a custom OpenAI-compatible endpoint. With a cloud provider, the text leaves your computer, so keep that option off for confidential work, or point it at a model you run yourself.
Which models and languages does Handy support?
Handy lets you choose between two families of local speech models:
- Whisper (OpenAI): Small, Medium, Turbo and Large, with GPU acceleration when available. Download sizes range from about 487 MB to 1.6 GB, according to the README.
- Parakeet V3 (NVIDIA): a “CPU-optimized model with excellent performance and automatic language detection”, useful on machines without a strong GPU.
You can also drop in your own Whisper GGML model files, which Handy detects automatically.
On languages, the two engines differ. NVIDIA’s model card lists 25 European languages for Parakeet TDT 0.6B v3, including English, French, German, Spanish, Italian and Portuguese. Whisper covers far more languages, but the Whisper project itself warns that its performance “varies widely depending on the language”. Handy does not publish its own language list, so test your languages before relying on it.
What features does Handy include?
For an app that promises “one tool, one job”, Handy has grown a useful settings panel. Based on its repository and recent release notes, you get:
- Custom words, to help the model spell names and jargon
- Filler-word removal (“um”, “uh”), refined in v0.9.8
- Transcript history, with copy to clipboard
- Microphone selection, audio feedback and an option to mute playback while recording
- CLI flags and a Raycast extension for remote control, model switching and dictionary management
- A Chinese script selector, added in v0.9.8
What it does not try to be: a file transcription tool, a meeting recorder or a voice-command system. It is a hotkey dictation app, and it stays focused on that.
What platforms does Handy run on, and what are its limits?
Handy’s v0.9.8 release ships installers for:
- macOS: Apple Silicon and Intel (DMG), also available through Homebrew
- Windows: x64 and ARM64 (EXE and MSI), also available through winget
- Linux: x64 and ARM64 (deb, rpm and AppImage)
The README does not list minimum OS versions. It is also refreshingly honest about known issues:
- Wayland on Linux: support is limited. You need
wtypeordotoolfor text input, and on Ubuntu 26.04ydotoolis required becausewtypedoes not work. - Linux overlay: the recording overlay is disabled by default because it can steal focus and break pasting.
- macOS: Bluetooth headset microphones can lower playback quality while recording, and
fn/Globe shortcuts only work on Apple keyboards. - Clipboard pasting: under load, Handy can restore your previous clipboard too early; a “Reliable Paste” beta option exists on macOS and Windows.
None of these is a dealbreaker, but together they show the trade-off of a free community project: you may need to troubleshoot yourself.
Handy vs Weesper: how do they compare?
Both apps run Whisper locally, use a global shortcut and insert text where you type. The difference is in the model of the product: Handy is a free community project; Weesper Neon Flow is a commercial app with support and no cloud features at all.
| Weesper Neon Flow | Handy | |
|---|---|---|
| Platforms | macOS 13+, Windows 10 (19041+) / 11 | macOS, Windows, Linux |
| Licence | Commercial | MIT open source |
| Local engines | Whisper | Whisper, Parakeet V3 |
| Cloud features | None | Optional AI post-processing via cloud providers |
| Languages (vendor claims) | 55+ | Depends on model; Parakeet V3: 25 |
| Custom vocabulary | Yes | Yes (custom words) |
| Support | Vendor support and Help Center | Community, GitHub issues |
| Price | 15-day trial, then €5/mo, €45/yr or €99 lifetime | Free (donations welcome) |
Where Handy is stronger: it costs nothing, its code is auditable and forkable under MIT, it runs on Linux, and it offers Parakeet for CPU-only machines.
Where Weesper Neon Flow is stronger: there is no cloud feature to switch on by accident, setup does not involve choosing between model families or Wayland input tools, and you get vendor support and a Help Center when something breaks. If you are comparing other free and local options, our VoiceTypr review, our Spokenly review and our guide to open-source dictation tools on Linux cover the field, and the Weesper comparison hub lists head-to-head pages.
Want to compare the two on your own documents? Start the 15-day Weesper trial, no credit card required.
Free open-source app or paid offline tool: what is the real trade-off?
The trade-off is your time, not your privacy: both keep audio on your device. Handy proves that excellent local dictation no longer has to cost anything. The real question is time: who maintains the setup, and who fixes it when an update or a new OS version breaks pasting?
With Handy, you are the administrator. That suits developers and tinkerers, and it suits anyone who wants to read the code before trusting it with their voice.
With a paid tool, you pay to skip that work. For Weesper Neon Flow’s own plans, the cost over time is:
| Weesper Neon Flow plan | After 3 years | After 5 years |
|---|---|---|
| Monthly (€5/month) | €5 × 36 = €180 | €5 × 60 = €300 |
| Annual (€45/year) | €45 × 3 = €135 | €45 × 5 = €225 |
| Lifetime (€99 once) | €99 | €99 |
The lifetime licence breaks even after €99 ÷ €5 = 19.8 months against the monthly plan. Our guide to the best free voice dictation software explains when a free tool is enough and when it is not.
Who should choose Handy, and who should look for an alternative?
Choose Handy if you want a free, private dictation app and don’t mind a little setup. It is a particularly strong pick if you use Linux, switch between three operating systems, or want MIT-licensed code you can audit and adapt. A developer on a Fedora laptop who dictates commit messages and chat replies, for example, gets local dictation at no cost and can fix a Wayland quirk by reading the README.
Look for a Handy alternative if you need a supported product for professional work, prefer a setup with no cloud options to review, or want a vendor to answer when dictation stops working before a deadline. A lawyer dictating client notes on a MacBook, for instance, gains little from forkable code and a lot from an app with no cloud feature to audit and a support channel to contact.
If that sounds like you, try Weesper Neon Flow free for 15 days — no credit card — and dictate your own work with both apps side by side. Setup questions are covered in the Weesper Help Center.
Verdict
In our assessment, Handy is the best free open-source dictation app for anyone who wants local speech to text across Mac, Windows and Linux. It is genuinely free, its privacy promise is backed by public code, and its developers document its limits openly.
Its weaknesses are those of any community project: no formal support, rough edges on Linux Wayland, and an AI cleanup option that sends text to the cloud if you choose a cloud provider. For personal use, those are easy trades. For confidential professional work on Mac or Windows, a supported offline-only tool removes the setup and the doubt.