Free and open source, like Handy. Ours never needs a second hotkey.
Both can clean up your dictation with AI. EnviousWispr's polish runs on the same hotkey as transcription, automatically once enabled: Apple Intelligence is the default on macOS 26+, needing no download, and on earlier versions our own EG-1 model, no account or API key needed, is one setting away. Handy's cleanup is real too, and now ships a ready-made prompt, but it is off until you turn it on, needs an API key for its default provider unless you switch to a keyless one, and only fires on a second, separate hotkey.
Quick answer
Handy or EnviousWispr: which should you choose?
Both apps are free, open source, and can keep dictation on your device. Choose Handy if you need Windows or Linux, want a large model catalog, prefer its permissive MIT license, or need to replay saved audio. Choose EnviousWispr if you want a curated Mac setup where one hotkey records, transcribes, polishes, and pastes, with Escape Recovery when a dictation is canceled by mistake. Handy is the better fit for people who enjoy choosing engines and hardware settings. EnviousWispr is the better fit for people who want local AI polish built into a simpler daily workflow. Both are credible choices for private speech-to-text. The deciding question is whether you want broad control across operating systems or a Mac-first flow with fewer switches.
Still deciding? See the best Mac dictation apps by use case or browse all 17 dictation app comparisons.
Feature comparison
An honest look at how the two tools stack up across the dimensions that matter most.
| EnviousWispr | Handy | |
|---|---|---|
| Price | Free. No subscription, no word limits. | Free. No subscription, no word limits. |
| Account required | No | No |
| Source code | Open source (GPLv3, copyleft) | Open source (MIT, permissive) |
| Speech engines | Parakeet TDT v3 + WhisperKit (dual engines, on-device) | Dozens of local choices across Parakeet, Nemotron, Canary, Cohere, Whisper, Voxtral, Qwen3 ASR, FunASR, Granite, Moonshine, SenseVoice, GigaAM, Breeze, and MedASR |
| Model choice | Parakeet by default; switch to WhisperKit in Settings for other languages | Extensive: filter the downloadable catalog by language, streaming, translation, and model family |
| Audio leaves your Mac | Never. If you enable cloud AI polish, your transcript, custom words, and the app you're dictating into are sent, never audio. | Never. No cloud engine documented. |
| Works offline | Yes (after models download) | Yes (after models download) |
| AI polish (cleanup after transcription) | EG-1, Apple Intelligence, or a local Ollama model on-device; OpenAI, Claude, or Gemini with your own key, or Ollama's hosted models with your Ollama sign-in | Yes: OpenAI, OpenRouter, Anthropic, Groq, Cerebras, Apple Intelligence, Custom, Z.AI, or AWS Bedrock through Mantle |
| Default provider needs a download or an API key | No, on macOS 26+: Apple Intelligence is the default, needing neither, when it's turned on for your Mac. On macOS 14-25, deterministic cleanup (filler words like um and uh, numbers/dates/times for English) still runs automatically; AI-level polish needs picking EG-1, no account or API key required, just a one-time 2.9GB download | Yes: off until you turn it on; the default provider is OpenAI, which needs your own API key, unless you switch to Apple Intelligence or a local server |
| Needs a prompt to be written | No: no prompt involved at all | No: ships a ready-made default prompt; you can edit it or write your own |
| How polish is triggered | Same hotkey as transcription, applies automatically once turned on | A second, dedicated hotkey, separate from plain transcription; you choose per recording |
| Filler word removal | Yes (built-in, automatic, language-aware, runs before AI polish) | Handled by the shipped Post Process prompt when that separate workflow is used |
| Turns a spoken list into a real list, no prompt needed | Yes (EG-1, 83% of the time in our own testing) | Depends on the selected prompt and provider; custom prompts are supported |
| Live preview while speaking | Yes: off by default, one setting to turn on | With some models: overlay is on by default, but only shows live text with a compatible Whisper model; falls back to a waveform with Parakeet |
| Recover a cancelled dictation | Yes (Escape Recovery, opt-in, keybind/triple-press cancel only, 24-hour History) | Not documented |
| Custom vocabulary | Yes (names, terms, jargon; imports from 8 other apps, Handy included) | Yes (built into Advanced settings; Raycast integration also exists) |
| Transcription latency | 0.61s median; ~1.65s with on-device AI polish | Not published |
| Clipboard preservation | Yes (clipboard saved before paste, restored after) | Configurable: leave the clipboard untouched or keep a transcript copy; paste with Command-V or choose no paste |
| Where transcription runs | Apple Neural Engine (Parakeet) or Apple GPU (WhisperKit), Apple Silicon only | Yes (Whisper models, on Intel/AMD/NVIDIA GPUs on Windows and Linux) |
| Custom model support | No: curated engines only | Yes: auto-discovers custom GGUF models from Hugging Face |
| Multi-language | 25 European (Parakeet), 99+ via WhisperKit | Broad via Whisper; automatic detection on Parakeet V3 |
| Platform support | macOS (Apple Silicon only) | macOS (Intel + Apple Silicon), Windows, Linux |
| Text lands in the right app | Yes (target app reactivation via Accessibility API, three-tier paste fallback) | Pastes in the app with focus when you release the hotkey |
Handy claims sourced from handy.computer, handy.computer/docs/post-processing, github.com/cjpais/Handy (README, Wiki, and source code), the GitHub API, and Handy v0.9.5 opened directly. EnviousWispr latency from production PostHog data on Apple Silicon Macs. Competitor claims last verified: 2026-08-21.
Both are free and open source. Here is what EnviousWispr adds
Handy gets the hard part right: fast, private, on-device transcription at no cost, and it can run AI cleanup too. The difference is how much work that cleanup takes to turn on.
Polish runs on the same hotkey as transcription, automatically once it is on. On macOS 26+, Apple Intelligence is the default provider, needing no download. On earlier versions, deterministic cleanup (filler words like um and uh, spoken numbers/dates/times when dictating in English) still runs automatically, and EG-1, our own model fine-tuned for dictation, is one setting away for the harder cases: a one-time 2.9GB download, no account or API key. Handy's cleanup is real and now ships a ready-made prompt too, but it is off until you turn it on, always asks you to enable it and supply an API key or switch to a keyless option, and fires on a second, separate hotkey.
Live Preview shows your words in the recording pill while you talk, so you always know you are being heard. It is off by default; turn it on in Settings, Record, Live Preview, and pick an engine: Apple's needs macOS 26, or Universal works from macOS 14 with a one-time 217MB download. Your finished text is still decoded from the whole recording after you stop, so nothing is lost waiting for the preview to catch up. Handy has a live overlay too, on by default, though it only shows live text with certain Whisper models and falls back to a waveform with Parakeet.
Escape Recovery keeps a dictation you cancel with your keybind or a triple press, instead of discarding it instantly. The app finishes transcribing and polishing, then offers an Undo button, and the text waits in your History for 24 hours if you let it fade. Off by default; you choose to turn it on. The Cancel button in the recording bar always discards immediately, with or without it on.
Measured in production on Apple Silicon Macs. Parakeet TDT runs on the Neural Engine specifically tuned for fast, accurate dictation, so text appears almost instantly after you stop talking.
Built-in filler-word removal runs before AI polish, and it is language-aware: with your language locked to German, Dutch, Danish, or Norwegian in Settings, it keeps real words like "er" and "um" instead of stripping them as English filler.
Smart Import reads your custom vocabulary straight out of Handy's own files, without changing them, and shows you every word before anything joins your library. No retyping.
Where does your voice go?
This is the rare comparison where both apps earn full marks by default. The difference only shows up if you choose to turn on an optional feature. For a deeper dive, read on-device vs cloud dictation privacy.
Parakeet TDT: built for speed
EnviousWispr uses NVIDIA Parakeet TDT v3, a multilingual ASR model covering 25 European languages that runs natively on the Apple Neural Engine. No network dependency, no server queues.
Based on production data from Apple Silicon Macs. Results vary by hardware and settings. Handy does not publish comparable latency figures.
What the feature table does not show
Both apps solve on-device transcription well. The real difference is in the choices each project made about what to build next.
Handy gives you a real menu: Whisper Small through Large with GPU acceleration, Parakeet V2 or V3, and it can auto-discover custom GGUF models you download from Hugging Face. If you run Handy on a Windows machine with a discrete GPU, that flexibility is genuinely useful, and it is one of the clearest reasons to pick Handy over EnviousWispr.
EnviousWispr keeps it to two: Parakeet TDT v3 is the default, tuned for speed and English/European accuracy. If you dictate in a language it doesn't cover, switch to WhisperKit under Settings → Transcription, a one-time toggle plus a roughly 1.5GB download the first time. The choice is manual, not automatic; there just isn't a size ladder to climb.
Handy's Post-Processing feature is real: a dedicated hotkey sends your transcript to a provider of your choosing, OpenAI, Anthropic, OpenRouter, Groq, Cerebras, Z.AI, Apple Intelligence on-device, or a local server, along with a prompt. It ships one ready-made prompt out of the box, called Improve Transcriptions, so you are not required to write your own to get started; you can also edit it or add your own. The feature itself is off until you turn it on; the default provider is OpenAI, which needs your own API key unless you switch to Apple Intelligence or a local server, and it fires on a separate hotkey from plain transcription. Whether it turns a rambling sentence into a real list, or reliably keeps the right half of a self-correction, depends on which prompt and provider you end up using.
EnviousWispr's polish runs on the same hotkey as transcription, automatically once it is on, and never needs a second hotkey the way Handy's does. Deterministic filler-word removal (um, uh, hmm, and similar) runs first and always, whatever language you're dictating in; spoken numbers, dates, and times convert to digits automatically too, when you're dictating in English. EG-1, our own model fine-tuned specifically for dictation, then handles the harder cases: turning a spoken list into a real one, keeping only what you meant when you talk yourself into a correction mid-sentence. That LLM step is guarded against hallucination: short transcripts skip the model entirely, and output beyond roughly three times the input's length (with a floor around 200 characters, so a short dictation isn't flagged the instant polish adds a sentence) is rejected as probable fabrication and the raw transcript is used instead.
In our own testing on the latest release, EG-1 builds a real bulleted list out of a spoken list 83% of the time (up from under 1%), and correctly keeps only what you meant when you self-correct mid-sentence 78% of the time. That is a measured result from a model built and tested for this one job. Handy's cleanup quality with its default prompt, or your own, has not been benchmarked publicly, and will vary further if you write your own instructions or switch providers.
Choose Handy if any of this matters more to you
Handy is a genuinely good, free, open source project, and it does some things EnviousWispr does not. It may be the better fit here:
Handy runs on macOS (Intel and Apple Silicon), Windows, and Linux from one codebase. EnviousWispr is Apple Silicon Mac only. If your dictation needs to follow you across operating systems, Handy already does that.
Handy's live catalog covered more than a dozen model families, with filters for language, streaming, and translation. EnviousWispr offers Parakeet or WhisperKit, which is simpler but leaves far less to tune.
Handy is MIT licensed, which lets anyone fold the code into a closed-source or commercial product without releasing their changes. EnviousWispr's GPLv3 is copyleft: derivative works you distribute must also be open under GPLv3.
Handy has over 30,000 stars on GitHub as of August 2026, more than most dictation apps we compare against, us included, along with an active contributor base and issue tracker.
Handy's Post Process list includes OpenAI, OpenRouter, Anthropic, Groq, Cerebras, Apple Intelligence, Custom, Z.AI, and AWS Bedrock through Mantle. EnviousWispr's cloud polish covers OpenAI, Claude, Gemini, and hosted Ollama. Handy's provider menu is wider.
Handy's History keeps an audio player alongside every saved entry, so you can play back what you actually said. EnviousWispr discards audio after every dictation by design, even in History, so this is not something we offer.
If you want a large local model catalog, detailed hardware controls, or Windows and Linux support, Handy is a strong pick. If you want a smaller curated setup with one hotkey for dictation and cleanup, give EnviousWispr a try. You can always use both.
Common questions
Yes. Both EnviousWispr and Handy are completely free, with no subscription, no word limits, and no account required. Both are also open source, so you can read the code that runs on your Mac.
Both transcribe on-device with no cloud dependency for speech recognition, and both can run AI cleanup afterward. The difference is how that cleanup works. EnviousWispr's polish runs on the same hotkey as transcription, automatically once it is on, and never needs a second hotkey: on macOS 26+, Apple Intelligence is the default provider and needs no download, when it's turned on for your Mac; on earlier versions, deterministic cleanup (removing filler words like um and uh, and converting spoken numbers, dates, and times to digits when dictating in English) still runs automatically, and EG-1, our own model, adds the harder cases like formatting spoken lists once you pick it, no account or API key, after a one-time 2.9GB download. EnviousWispr also adds a Live Preview of your words as you speak and Escape Recovery for accidentally cancelled dictations. Handy's cleanup, called Post-Processing, is genuinely capable, supporting more cloud providers than we do (OpenAI, Anthropic, OpenRouter, Groq, Cerebras, Z.AI, or Apple Intelligence), and it now ships a ready-made default prompt, so you are not required to write your own. But it is off until you turn it on, always asks you to enable it and supply an API key or switch to a keyless option, and fires on a separate dedicated hotkey from plain transcription. Handy in turn runs on Windows and Linux in addition to Mac, and lets you choose among several Whisper model sizes for a speed-versus-accuracy trade-off; EnviousWispr is Apple Silicon Mac only.
Yes, it does. Handy's Post-Processing feature sends your transcript to a provider of your choice (OpenAI, Anthropic, OpenRouter, Groq, Cerebras, Z.AI, Apple Intelligence on-device, or a local server) along with a prompt. Handy ships one ready-made default prompt, called Improve Transcriptions, so you don't have to write your own to get started, though you can edit it or add your own. The feature is off until you turn it on, and it fires on a separate dedicated hotkey from plain transcription, so you choose per recording whether to use it. EnviousWispr's polish runs on the same hotkey as transcription and needs no prompt to write: on macOS 26+ that's Apple Intelligence, the default provider needing no download, and on earlier versions EG-1, our own model already fine-tuned for dictation cleanup, is one setting away with no account or API key needed.
Yes. After the one-time model download, transcription runs entirely on-device with no internet required. AI polish can also run offline using EG-1 (our own model, macOS 14+), Apple Intelligence (macOS 26+), or Ollama with a local model. Cloud providers like OpenAI, Claude, or Gemini are optional and require your own API key.
Yes, though under a different license. EnviousWispr is licensed under the GNU General Public License v3 (GPLv3), a copyleft license: anyone can read, build, and modify the code, but a distributed derivative must also be open under GPLv3. Handy is licensed under the MIT license, which is more permissive and lets anyone fold the code into a closed-source or commercial product. Both are genuinely open source; the licenses just make different trade-offs.
Yes. EnviousWispr's Smart Import reads your custom vocabulary directly out of Handy, without changing Handy's own files, and shows you every word it found before anything joins your library. It also imports from Wispr Flow, Superwhisper, FluidVoice, TypeWhisper, Spokenly, Juno, and Vox.
Not today. EnviousWispr runs on Apple Silicon Macs only (macOS 14 Sonoma or later). Handy supports macOS (Intel and Apple Silicon), Windows, and Linux, which is a real advantage if you need one dictation tool across multiple operating systems.
Any Mac with Apple Silicon (M1 or later) running macOS 14 Sonoma or newer. The Neural Engine on Apple Silicon is what makes on-device transcription fast. Handy also supports Intel Macs, which EnviousWispr does not.
EnviousWispr runs NVIDIA Parakeet TDT v3 and WhisperKit. Handy 0.9.5 showed dozens of downloadable choices across Parakeet, Nemotron, Canary, Cohere, Whisper, Voxtral, Qwen3 ASR, FunASR, Granite, Moonshine, SenseVoice, GigaAM, Breeze, and MedASR.
Yes. Download EnviousWispr, set your keybind, and Smart Import can bring your Handy custom words across in one pass. Handy's own filler-word list, prompts, and tuning settings stay behind, since those are Handy-specific settings rather than words to keep, not a data-loss gap. Both apps work in any text field on macOS, so nothing about your workflow has to change.
Free and open source. Never a second hotkey.
Free to download. No account required. AI polish runs on your one hotkey, automatic on macOS 26+ and one setting away everywhere else.