Stays out of the way
One pill at the bottom of the screen while you talk, and nothing else. No window opens, no focus is stolen, no modal interrupts you.
Dictation for macOS · MIT licensed · bring your own key
Speak anywhere.
VoxFlow writes it for you.
Hold ⌥ ⌃, say the thing, let go. Apple Speech transcribes it on your Mac, an AI pass turns the rambling into a sentence, and the text lands at your cursor in whatever app you were already in.
macOS 26+ · Apple Silicon · open the DMG, drag to Applications
⌥⌃ held
audio in
−1.2 s silence trimmed
transcript
um so can you uh check whether the the cache is warm before we ship
rewrite
Can you check whether the cache is warm before we ship?
only when the text needs itHow it works
Five stages between your voice and your cursor.
Everything on the hot path is Rust — the hotkey tap, audio capture, the voice detector, the state machine. The interface only draws what already happened.
- 01
Hold the keys
Hold Option+Ctrl anywhere in macOS. A small pill fades in at the bottom of the screen. Nothing takes focus, nothing covers what you were reading.
- 02
Speak
Audio is captured through CoreAudio and gated by a voice activity detector, so the silence at either end never reaches the model.
- 03
Apple Speech, on device
Transcription runs on your Mac through Apple's SpeechAnalyzer. When it comes back unsure, the audio is re-run through local Whisper with Metal acceleration. Either way it does not leave the machine — the fallback model downloads once, on first use.
- 04
An AI pass cleans it up
The raw transcript goes through a rewrite pass that drops fillers, fixes grammar, and returns a sentence you would have typed. On by default, not a paid add-on.
- 05
It lands at your cursor
The finished text is pasted into whatever app had focus — editor, terminal, browser, chat. VoxFlow puts the text on the clipboard, sends Cmd+V, and then puts your old clipboard back.
What you get
Built to feel like part of the OS.
Built to stay out of the latency budget
VoxFlow targets an 80 ms pill and text inside 1.5 s for a short phrase. The hot path is Rust; the interface only renders.
Your audio stays home
Apple Speech runs on device, with local Whisper as an automatic confidence fallback. Only eligible transcript text can reach an optional rewrite provider.
Works where you write
Insertion is system-wide, so the same hotkey works in Cursor, VS Code, Slack, Gmail, Notion, Obsidian, browsers, and terminals — anywhere Cmd+V works.
Keys live in the Keychain
Provider keys are stored through the macOS Keychain, never in a plaintext settings file. Local history is a SQLite file you own.
Different voice per app
Per-app profiles pick the output mode for whatever has focus — Markdown in Obsidian, Terminal-safe in iTerm, Code-preserve in Cursor, Email in Mail.
Learns the words you use
Add preferred spellings and VoxFlow biases both speech engines toward them, so it writes handoff instead of hand of. Fix a word by hand and it offers to remember it.
Spends the expensive model only when it matters
Hybrid routing keeps short, clean audio on the cheap model and sends anything long or noisy to the accurate one, then drops to the cheap tier once you hit your cap.
Costs you can see
VoxFlow tracks what each dictation costs, projects the month, and warns you at 50, 80, and 100 percent of a cap you set.
Inside the app
Settings that match how you actually work.
VoxFlow is one pill while you talk. Everything else lives in a settings window you open when you need it, and not before.
Bring your own key
Pick where the work happens.
Transcription is local by default. The rewrite pass talks to any OpenAI-compatible endpoint, so you can route it to Groq, to OpenAI, to a machine in your own house, or turn it off entirely.
Fully local
$0 per month
Apple Speech with local Whisper fallback, cleanup off. Nothing leaves the machine.
Local + cloud rewrite
cents per month
On-device transcription, with Groq handling only adaptive cleanup on short text.
Your own server
$0 marginal
Point VoxFlow at any OpenAI-compatible server you run — a home box over Tailscale, say.
Costs depend on your own provider and usage. VoxFlow tracks each dictation and projects your month so nothing arrives as a surprise. Seeprovider keys for setup.
Privacy
Your voice never has to leave the machine.
- Audio is transcribed on device by Apple Speech or local Whisper — no upload, no account, no transcript telemetry.
- Provider keys are stored in the macOS Keychain, not in a settings file.
- Transcript history is a local SQLite database you can read or delete yourself.
- Turn the rewrite pass off and VoxFlow makes no network calls at all.
- The whole thing is MIT licensed, so you can read every line before trusting it.
Pricing
The app is free.
Never for the minutes. No account. MIT licensed. You only pay a provider if you turn the rewrite pass on and point it at one.
$0
what VoxFlow charges. Forever.
Included
Local Whisper, the hotkey, the overlay, rewrite, history, as many providers as you want. The whole app.
You may still pay
Only Groq, OpenAI, or a server you run, and only if you send text there. Turn rewrite off and nothing leaves the machine.
Questions
The things people ask first.
Does my voice get sent to the cloud?
No. Transcription runs on your Mac — Apple’s SpeechAnalyzer first, and local Whisper when it is unsure. The Whisper fallback model is downloaded once, on first use. Only the resulting text can reach a rewrite provider, and you can turn the rewrite pass off to keep everything on device.
Is VoxFlow a subscription?
No. The app is free. There is no VoxFlow subscription and no account. If you send the rewrite pass to Groq or OpenAI, you pay that provider directly for what you use.
Which apps does it work in?
Any app that accepts text. VoxFlow copies the result to the clipboard and sends Cmd+V, then restores what you had on the clipboard before — so it works in editors, terminals, browsers, chat apps, and email. Accessibility permission is what lets VoxFlow watch for the hotkey.
What is the hotkey?
Option+Ctrl by default, held or toggled. It uses a CGEventTap rather than a letter-key global shortcut, so it does not collide with shortcuts inside the app you are using.
Which platforms are supported?
macOS 26 or later on Apple Silicon.
How is this different from Wispr Flow?
Same idea, different economics. VoxFlow is free. Transcription stays on your Mac, keys live in the Keychain, and you pay a provider only if you choose to. Wispr Flow charges a monthly subscription for the minutes.