Changelog
What shipped.
Dated and honest, newest first. Everything listed here works today, on a machine you own. "Coming" means being built, not being teased.
27 July 2026Windows
Windows grew up: a real installer.
- Download one FastspokenSetup file, run it, and you are dictating. Start Menu and desktop shortcuts, an optional start-at-sign-in, and a proper uninstaller. No zip, no digging for the right file.
- The app is its own program now: Fastspoken in Task Manager, Fastspoken on the taskbar, the mouth on every window. No more borrowed plumage.
- Onboarding opens the moment the install finishes and greets you with the wordmark, matching Android.
- Prefer no installer? The portable zip is still on the download page.
26 July 2026WindowsAndroidEngine
Windows learned to get out of your way, and your paragraphs learned to finish themselves.
- Paragraph settle, on both platforms: pause for a breath and the thought you just finished quietly re-reads itself in full, so commas and sentence flow land the way a person would write them. Runs on your device, free tier, while you keep talking. Android 0.16 and Windows 2.0.2 carry it today.
- Raw dictation tidies itself as it lands: stray mid-sentence capitals fixed, lone "i" capitalized, spacing repaired. Your words are never changed, only their dress.
- The pill grew manners: a taskbar button, a hover minimize, a right-click menu with Hide and Quit, and double-tap Esc to duck it. It stays on top when you want it and disappears when you don't.
- Download moved to a versioned build (2.0.2) so updates arrive the moment they ship.
24 July 2026Windows
Fastspoken came to Windows.
- A small floating pill: hold Right Ctrl and talk, text types into whatever app you are in. Unzip and run, no installer, no admin, no account.
- The same on-device engine as Android: your voice never leaves your machine on the free plan.
- Raw | Cleaned switch: verbatim always free, polished is Pro.
9 July 2026Web
The site grew a whole marketing wing.
- Mobile navigation: a proper menu, with your account one tap away when signed in.
- New pages: Why Pro, Private by architecture, vs Wispr Flow, the demo, use cases, and this changelog.
- Free tools: a typing speed test with a talking test to match.
- Every speed claim on the site re-sourced to published research, labeled as averages, with the studies named.
8 July 2026WebAndroid
fastspoken.com went live. The dictionary went zero-knowledge.
- The site, accounts, and the "your book" dashboard: lifetime words, time saved, streak, and a warm heatmap, mirrored from the app's on-device ledger as counts only.
- End-to-end encrypted dictionary and snippets sync (Pro): sealed on your device with a passphrase only you know. We proved to ourselves we can't read it, then shipped it.
- Cloud polish API (Pro): filler removed, self-corrections resolved, process-then-discard.
- Android: the on-device engine passed its speed gate, decoding chunks in 70 to 500 milliseconds, and on-device dictations now get the same polish pass.
7 July 2026AndroidEngine
Fastspoken got its name, its look, and its engine.
- The product became Fastspoken. Warm paper, ink, one coral moment: the design language landed across the app.
- New four-step onboarding, calm segmented controls, and the bubble's choreography: ink to coral to amber and back to calm.
- The on-device engine was chosen the hard way: benchmarked on the hard split of a standard speech set (LibriSpeech test-other) against the cloud-grade model it replaced. It scored 96.4% to the cloud model's 95.7%, on your phone, and got the job.
- Dictation moved fully to the keyboard layer, which is why it works in apps that block accessibility tools, Samsung Notes included.
- Silence handling: a voice-activity gate so quiet rooms produce nothing instead of hallucinated text.
6 July 2026Android
The bubble was born.
- A floating bubble that types what you say into any app: tap, talk, and whole sentences land at your cursor at every natural pause.
- Adaptive mic thresholds, so quiet voices and noisy rooms both work without tuning.
- A rescue net: if a landing is ever blocked, the full take arrives at the end. Words are never silently lost.
- Survives reboots, battery management, and network blips.
4 July 2026Engine
First words.
- The dictation pipeline came alive: speech to text with an AI cleanup pass, a personal dictionary that primes recognition, and snippets that expand as you speak.
- The privacy laws were written before the product had a name: the live beat never requires a network, and servers never store what you say.
Android and Windows are live, both in early access. iOS is planned and not built yet. It will appear here when it is real, not before.