FREE Voice-to-text for Mac · v0.10.0

Speak. Paste. Done.

A free Mac alternative to subscription dictation apps. Hold a hotkey, talk, release. Whisper and Parakeet run entirely on your Mac. No cloud, no accounts, no audio leaves your machine.

macOS 15+ · Apple Silicon · Free forever · Built by Enginecy · Windows app coming soon
Slack — #design
VOIXE · LISTENING
HOLD
Hey team — quick update on the new onboarding flow. I think we should swap the second step for the demo video and push the email capture to the end
Works everywhere on macOS
Slack Messages Mail VS Code Notion Notes Linear Chrome Xcode Terminal

Mac now. Windows next.

Voixe is live for Apple Silicon Macs today. A Windows app is already in progress and planned for the next few days, with the same core promise: quick dictation that stays focused on the workflow.

Mac app available Windows app coming soon

Want Windows access and release notes?

Download stays open and free. Leave your email only if you want launch updates, new builds, and useful release notes.

No account required. Email is optional.
Voixe in the macOS menu bar, with the dictation indicator floating near the cursor.
Voixe lives in your menu bar. Open it to switch microphone, model, or theme without leaving the keyboard.

Built for dictation, not surveillance.

Voixe is a single hotkey, on-device models, and a strict rule: your voice never leaves the machine. Everything else is a polish layer on top.

On-device transcription

Whisper and Parakeet run via Core ML on Apple Silicon. The audio is processed locally; there is no server to send it to.

One hotkey, every app

Default is Option. Hold-to-talk or double-tap-to-toggle. Works in Slack, Notes, Xcode, Mail. Wherever the cursor is, the text appears.

Punctuation, two ways

Pause naturally and the model adds the commas. Or say "comma", "question mark", "new paragraph" out loud. Voixe ships with the full Apple Dictation vocabulary pre-wired.

Local Refine pass

Like the cleanup layer people love in Whisper Flow: a local LLM removes fillers, fixes punctuation, follows spoken commands, and matches the tone of the active app. Included free.

Multilingual by default

Parakeet TDT v3 covers 25+ languages out of the box. Whisper Large v3 if you need the long tail. Switch in two clicks.

Faster than typing

Sub-100ms transcription latency on M-series Macs. By the time you release the hotkey, the text is already pasted.

Designed for people who think faster than they type.

Press, speak, release.

One muscle memory. The waveform tells you it's listening. Releasing pastes wherever the cursor was when you pressed.

Hold Speak Paste

Refine reads what you meant.

Opt in to a tiny local LLM. It strips fillers and matches the register of the app you're typing into. Off by default. Your call.

Raw so like um can you grab me a coffee uh actually make it a tea
Refined Can you grab me a tea?

Your voice stays here.

No network call. No upload. No optional toggle in some buried setting. The audio is a temporary file that exists long enough to transcribe and then it's gone.

Audio → server — denied —
Audio → disk (temp) scoped, ~80ms
Transcript → clipboard paste, then gone

Pick your model.

Start instantly with built-in Apple Speech, no download. Parakeet is the fast, multilingual default; Whisper Large is state-of-the-art on hard audio. Swap them in Settings.

Parakeet TDT v3
1.5 GB · 25+ langs · ~80ms
Whisper Large v3
3.0 GB · 99 langs · ~280ms
Apple Speech
Built-in · 0 MB · zero download

All-day Mac dictation, without the subscription.

If you found Voixe while comparing Wispr Flow, Whisper Flow, Superwhisper, or Spokenly, this is the short version: Voixe is Mac-first, private, and 100% free.

Free forever
No pro tier Local Refine
  • On-device transcription with Parakeet and Whisper models.
  • Local Refine cleans filler words, punctuation, and tone like a Whisper Flow-style polish pass.
  • No account, BYOK setup, cloud workspace, usage cap, or paid pro tier.
  • Hold-to-talk and double-tap recording built for daily Mac work.
  • Built by Enginecy as a public proof of useful AI product engineering.
Broad voice assistants Free / Pro / BYOK
  • Often optimized for cross-platform use, agents, commands, and broader workflows.
  • Free tiers can be useful, but advanced features may depend on Pro plans or API keys.
  • More surface area if you want mobile, Windows, file transcription, or automation modes.
  • Best when you want a general voice layer, not a tiny Mac dictation tool.
Apple Dictation Built in
  • Free and already on every Mac.
  • Great for short bursts with no extra setup.
  • Less control over models, hotkey behavior, and post-processing.
  • Not designed as a focused workflow tool for power users.
Use Voixe when you want private Mac dictation. Talk to Enginecy when your business needs custom AI tools, internal apps, or automation with the same practical bias. Work with Enginecy

One hotkey. Four quiet moves.

Install

Drag Voixe to Applications.

Download the notarized DMG, move Voixe into Applications, and launch it from your menu bar.

DMG to Apps to Menu bar
Permissions

Grant only what the workflow needs.

Microphone captures your voice. Input Monitoring hears the hotkey. Accessibility pastes the final text.

Mic Hotkey Paste
Dictate

Hold Option and talk.

Press and hold for quick capture, or double-tap to lock recording while you think out loud.

Hold
Paste

Release. Voixe pastes.

Parakeet or Whisper transcribes locally, optional Refine cleans it up, and the text lands at your cursor.

On-device Refine optional ⌘V

Questions people actually ask.

Is the audio really not uploaded anywhere?
Correct. The transcription model runs in-process on your Mac via Core ML. Voixe's sandbox entitlements explicitly forbid outgoing network connections for audio. The only network calls Voixe makes are to download a model (one time, from Hugging Face) and to check for app updates (Sparkle, against voixe.enginecy.io/appcast.xml). Both are opt-out-able.
Why does Voixe need Input Monitoring?
So the global hotkey works in every app, not just when Voixe is in the foreground. macOS requires Input Monitoring permission to register a key listener that runs across all apps. Voixe only watches for your configured hotkey; every other keystroke is ignored.
Which Macs does it run on?
Apple Silicon (M1 and later), macOS 15 (Sequoia) or newer. Intel Macs aren't supported. The on-device models rely on Apple's Neural Engine for the latency you'd actually want.
How is this different from macOS's built-in dictation?
macOS Dictation is good but the hotkey behavior is limited and the model isn't tunable. Voixe gives you press-and-hold or double-tap modes, lets you pick between Parakeet (fast, multilingual) and Whisper (accurate, 99 languages), and adds an optional LLM "Refine" pass that fixes punctuation and removes filler words.
Is Voixe really free?
Yes. There's no account, no usage tier, no paid pro plan hiding behind a feature toggle. Enginecy ships Voixe as a calling-card for the engineering work we do for businesses. See the section below.
Is Voixe a free alternative to Whisper Flow or Wispr Flow?
For Mac users who mainly want private voice-to-text in every app, yes. Voixe focuses on free, on-device Mac dictation instead of subscriptions, cross-platform sync, or team account features. Refine gives you the same kind of cleanup layer: filler removal, punctuation repair, command following, and tone matching after transcription.
How is Voixe different from Spokenly?
Spokenly is a broader voice productivity tool. Voixe is intentionally narrower: fast Mac dictation, local Parakeet and Whisper transcription, optional local Refine, no account, no usage cap, no BYOK requirement, and no paid pro tier. If you want cross-platform voice workflows and assistant-style commands, Spokenly may make sense. If you want a small Mac app that stays free and pastes clean text anywhere, Voixe is built for that.
What is the "Refine" pass exactly?
A small local language model (Qwen 0.5B–3B or Llama 3.2 1B, your pick) that runs after transcription and cleans up the text. It removes "um", fixes punctuation, and adapts tone to the app you're pasting into (terse in Xcode, conversational in Slack). It's strictly opt-in: if you don't enable it, the raw transcription is what gets pasted.
From Enginecy

Voixe is the demo. Your workflow is the work.

Voixe shows how we build at Enginecy: find the repetitive work, design the smallest useful workflow, then ship the product-grade AI around it.

For business owners, that becomes custom software, AI agents, and automations that remove manual steps without adding another dashboard nobody wants to use.

Internal tools that replace fragile spreadsheets. AI workflows that fit how your team already works. Fast prototypes that can become real production systems.
See what we build
Voixe Settings, Transcription Model tab. Parakeet TDT v3 selected, with Whisper Small, Medium, and Large v3 as alternatives.