Hold a key. Talk.
It's already typed.
Speech-to-text and voice typing that works in every app on your Mac and never sends a byte anywhere. Your voice is transcribed on the Neural Engine in under a second, cleaned up, and typed straight at your cursor.
The bar sits above your dock while you speak, and gets out of the way when you stop.
Three seconds of setup per sentence.
Hold fn anywhere
In Mail, Slack, your editor, a browser field. LocalFlow does not care which app has focus — it works with all of them.
Say what you mean
Stumble, correct yourself, use filler words. That is what the cleanup is for. Press esc to throw one away.
Let go
The transcript is cleaned up and typed at your cursor. Under a second on Apple Silicon, with nothing sent anywhere.
What the cleanup actually removes.
Pick an example to see what happens between the microphone and the text field. This runs in your browser — it illustrates the pipeline rather than transcribing live.
Native, frugal, self-contained.
Written in Swift and SwiftUI. A 2.6 MB menu bar agent with no runtime dependencies beyond macOS itself.
Neural Engine inference
WhisperKit through CoreML, running on Apple Silicon's dedicated hardware rather than burning a CPU core or a datacentre GPU.
Nothing leaves your Mac
No audio, transcript or usage data is ever transmitted. Recordings exist in memory for the length of a transcription and are then gone.
Types where your cursor is
Inserts at the caret through the macOS Accessibility APIs, so it works in Safari, Slack, VS Code, Notes, Terminal and everything else.
Cleanup you control
Filler words, self-corrections and punctuation are handled by rules you can switch off one by one. Nothing is rewritten behind your back.
Your choice of hotkey
Push-to-talk on fn, right ⌥ or the dictation key. Or toggle with fn+Space. Double-press for hands-free.
Local history & dictionary
Search everything you have dictated and re-insert it. Teach LocalFlow the names and jargon it should always spell your way.
Free to use. €15 if you want the smart parts.
One payment, no subscription, no account. Works on every Mac you own and every future update is included.
A complete dictation app. Not a trial, and not time-limited.
Download- Unlimited dictation — no word or time caps
- Rule-based cleanup, filler words and self-corrections
- Tiny and Base speech models
- 10 dictionary rules · last 25 transcripts
- Smart AI polish
- Whisper Small model
Everything in Free, plus the parts that take real work to maintain.
- Smart AI polish — a local language model cleans up the whole transcript
- Whisper Small — the most accurate tier, for technical vocabulary
- Per-app writing styles — formal in Mail, casual in chat, code-aware in editors
- Unlimited dictionary rules and full history
- Every Mac you own · all future updates
Payment handled by Stripe. No VAT is charged or shown (small-business rule, § 19 UStG). Your license key is verified offline — LocalFlow never checks in with a server, not even to confirm you paid.
Against the alternatives.
How LocalFlow compares as a Wispr Flow alternative or a Superwhisper alternative — same job, nothing leaves your Mac.
| Capability | LocalFlow | Wispr Flow | Superwhisper | macOS built-in |
|---|---|---|---|---|
| Where inference runs | On-device (ANE) | Cloud servers | Local / hybrid | Hybrid |
| Cost | Free, €15 once for Pro | $12–15 / month | $8 / month | Included |
| Latency | ~0.7s | 1.2–2.5s (network) | ~0.9s | Varies |
| Audio uploaded | None (0 bytes) | Yes | Depends on mode | Partly |
| Works offline | Always | No | Depends on mode | Partly |
| Account required | No | Yes | Yes | Apple ID |
| Source readable | Yes | No | No | No |
Questions worth asking.
Does audio really never leave my Mac?
Yes. There is no networking code in the transcription path. LocalFlow touches the internet exactly twice: when you download a speech model, and when you buy Pro. Pull the network cable and everything else keeps working — including your Pro license, which is verified offline.
What is the catch with the free version?
There isn't one. Free is unlimited dictation with the full rule-based cleanup and two of the three speech models. No word caps, no trial period, no nag screens. Pro adds the local language model polish, the most accurate Whisper tier and per-app styles.
The source is public. Can't I just remove the license check?
Technically, yes — and that is the price of publishing the source. Legally, no: LocalFlow is under the Elastic License 2.0, which allows you to read, fork, modify and build it, but not to remove or work around the license key. Practically: it is maintained by one person, and €15 once is what makes that sustainable. If you genuinely cannot afford it, use the free version with a clear conscience — it is a real product, not crippleware.
Which languages does it handle?
Whisper's multilingual models, so most major languages. The cleanup rules are tuned for German and English. Auto-detect copes well with mixing the two mid-sentence; naming a language explicitly helps on very short phrases.
Will macOS warn me when I open it?
No. Releases are signed with a Developer ID certificate and notarised by Apple, so LocalFlow opens like any other app — no warning, no right-click trick. If you would rather not trust a prebuilt binary, it compiles from source with one script.
Does it work on Intel Macs?
It runs, using the whisper.cpp backend instead of the Neural Engine. Expect noticeably higher latency than the sub-second figures here, which are measured on Apple Silicon.
Why does it need Accessibility access?
It is the only supported way for a macOS app to type into another app's text field. LocalFlow uses it to insert at your cursor and — if you leave the option on — to read the few characters before the cursor so it can match capitalisation. That context is never stored and never read from password fields.
I lost my license key.
It is in your purchase confirmation email from Stripe. The success page link in that email also still works — open it again and the key is there. Failing that, open an issue with your order number.