Press a key, speak naturally, text appears - anywhere on your Mac. Powered by Parakeet TDT on the Apple Neural Engine. Zero cloud. Zero subscription tax.
4x faster than typing
Raw voice input
um so basically I think we should uh refactor the the useEffect hook because its like causing a memory leak when the component unmounts and uh yeah we need to clean up the subscription
Clean output
Refactor the useEffect hook - it's causing a memory leak when the component unmounts. We need to clean up the subscription on cleanup.
Simple by design
From install to first word in under two minutes. Works in any Mac app you already use.
Press Fn or your custom hotkey to begin recording. A subtle indicator appears in your menu bar.
Talk naturally, including technical jargon. TypeScript, GraphQL, Kubernetes - spelled correctly, every time.
Text appears in your focused app. No copy-paste, no window switching. Release the key, your words are there.
Built for developers
Purpose-built for people who care about privacy and spend their day in terminals and editors.
All processing happens on your Mac. Zero network requests. Your audio never leaves your device - not even in anonymized form.
A built-in local LLM removes filler words and fixes punctuation - without rewriting your voice. Runs entirely on your Mac, no separate install.
Custom word lists prime the transcription engine. useEffect, gRPC, Zustand - spelled right, no training required.
iTerm, Obsidian, Chrome, Cursor, Slack - any Mac app. Three injection methods ensure compatibility across native and Electron apps.
Pure Swift. Single binary. Menu bar daemon. Under 200MB with model loaded. No Electron, no runtime, no background update checks.
Map dictation to any key combo. Push-to-talk on Fn, hands-free toggle on F5, or whatever fits your flow. System-reserved combos are flagged before you save.
Compatible with your stack
Where it shines
Dictation stops feeling awkward the moment the tool respects the text you're writing. Here's where Voxstr earns its keep.
Describe the change out loud and let Voxstr type it into the terminal, GitHub, or Linear. Developer vocabulary keeps useEffect, Dockerfile, and kubectl spelled correctly every time.
Stop context-switching to type a three-line reply. Hold the hotkey, speak, release - the message is already in the Slack input box, cleaned up and ready to send.
Capture fleeting thoughts at the speed you think them. Voxstr drops cleanly formatted markdown into Obsidian, Bear, or Apple Notes - perfect for meeting summaries, research snippets, and end-of-day journals.
Talk through a design doc or RFC out loud. Voxstr's local cleanup removes ums and false starts so your first draft actually reads like prose. Takes ten minutes instead of an hour.
How we stack up
The only local-first dictation tool built specifically for developers.
| Feature |
Voxstr
from $1/mo
|
WisprFlow
$14.99/mo
|
Superwhisper
$11.99/mo
|
OpenWhispr
Free (open source)
|
|---|---|---|---|---|
| Local processing | ✓ | Partial | Partial | ✓ |
| Developer vocabulary | ✓ | ✕ | Limited | ✕ |
| Open cleanup prompt | ✓ | ✕ | Limited | ✕ |
| Native macOS app | ✓ | ✓ | ✓ | ✕ |
| AI cleanup | Local (built-in) | Cloud | Cloud | ✕ |
| Works in all apps | ✓ | ✓ | ✓ | Limited |
| Lifetime purchase option | ✓ | ✕ | ✕ | N/A |
No subscription surprises
Start free. Stay if you love it. Own it forever if you want.
Flexible, cancel anytime
Best value for power users
Pay once, yours forever
3,000 words free • No credit card required • Cancel anytime
Verifiable privacy
Every claim is specific and verifiable. Don't take our word for it - check for yourself.
Transcription makes 0 outbound connections. Your audio stays in memory on your machine.
Verify: Open Little Snitch or Activity Monitor during transcription.
Audio is processed by the Apple Neural Engine via Parakeet TDT. Never touches a server, not even ours.
Verify: Check Activity Monitor for ANE usage during transcription.
No analytics, no crash reports containing audio, no usage tracking of any kind.
Verify: Monitor network with Wireshark or Charles Proxy.
Audio buffers are processed in memory and never written to disk. Transcription history is stored in your local app sandbox - delete it any time from Settings.
Verify: Inspect app sandbox with fs_usage during transcription.
License verification on first launch and periodically. That is the only outbound connection Voxstr ever makes.
Verify: Allow only license server in Little Snitch. Everything works.
Platform commitment
We build one app for one platform, and make it exceptional.
No Electron, no runtime. Pure Swift for instant response and minimal resource usage.
Transcription runs on dedicated ML hardware, not your CPU. Battery-friendly and fast.
Pure Swift binary, ~150MB. Models download once on first launch and stay on your Mac. No runtimes, no bundled browsers.
Always available, never in the way. No dock icon, no windows to manage.
Wave 1 beta launches in days. Reserve a spot and we'll send the build the moment it's signed.
macOS 14+ • Apple Silicon • 3,000 words free, no credit card
Questions