Frequently asked questions.

A desktop app that turns your voice into text in any application — press a hotkey, speak, release, your words appear where the cursor was.

Windows 10/11 and macOS 12+ on Apple Silicon (M1 or later). Both builds are first-class — same hotkey, same UI, same models. Intel Macs and Linux aren't supported.

Yes. The free tier is unlimited dictation on the two compact local Whisper models. No watermark, no 15-minute trial timer, no monthly word cap, no account required. Use it as long as you keep the app installed.

About 30 seconds. The installer is around 20 MB. The first model download (about 75 MB) happens automatically on first run.

Only for Pro. The free tier works without one. When you upgrade to Pro, the account ties your subscription to up to 3 devices.

Probably yes for the free tier. The compact models run on a regular CPU and finish a short phrase in 1–3 seconds. The bigger Pro models prefer a GPU but will still run on CPU — slower, but they work.

On your machine. We use whisper.cpp (the open-source C++ port of OpenAI's Whisper) running locally. Audio is processed in RAM and discarded the moment we have your text.

Correct. In the free tier, audio is processed entirely on your machine. In Pro, the bigger local models work the same way. We don't currently ship a cloud STT option — if we ever do, it will be a Pro-only opt-in, off by default.

A model is the AI that does the speech-to-text. Bigger models are more accurate but use more RAM and CPU/GPU. SnailText ships three tiers:

• · Compact (free) — fastest, fine for short messages and English
• · Advanced (Pro) — noticeably better on accents, technical terms, non-English
• · Maximum (Pro) — preferred for long-form, multilingual, demanding work

For short phrases (under 10 seconds): typically under a second on a GPU, 1–3 seconds on a modern CPU. The text appears after you finish recording — we don't stream it word-by-word as you speak.

Yes. After the initial model download, you don't need an internet connection. The free tier works fully offline. Pro license verification reaches out once a day to confirm your subscription, but everything else stays local.

Whisper recognises 100+ spoken languages with varying accuracy — English, German, Spanish, French, Italian, Portuguese, Russian, Polish, Turkish, Chinese, Japanese, Korean and others land in publication-grade territory. Rarer languages are usable but more error-prone. The Pro tier adds an alternative engine that supports 25+ European languages with very high accuracy.

No. Audio is held in RAM during recording, sent to the local model, and freed once the text is produced. Nothing is written to disk, nothing is sent to a server.

No. We don't have it. We can't train on data we don't have.

Yes — anonymous, opt-out. We use PostHog (EU-hosted) for product analytics: events covering app lifecycle, feature usage, errors. The events contain counts and event names, never the text you dictated, never your audio, never your filenames. Your distinct ID is a hash of a random device UUID — there's no way to map it back to you. Toggle it off in Settings → Privacy.

That's a local SQLite file on your machine — searchable history of what you dictated. It never leaves your device. You can delete individual entries, set a retention period, or wipe everything from Settings → Privacy.

The website uses Cloudflare Web Analytics (no cookies, no IP logging) and embeds Bunny Fonts (privacy-friendly, no Google tracking). We don't drop tracking cookies, and we don't track you across the web.

The full privacy policy is at /privacy/. For a Data Processing Agreement (e.g. for company use), email [email protected] — we'll work with you on a template.

$7.49 / month or $89 / year (about $7.42/month annualised). The yearly price is for users committing for a year, not a discount-the-honest-way trick. Full comparison and the billing toggle live at /pricing/.

Our backend is built around subscription verification. Adding lifetime would mean a real refactor we haven't done. We'd rather start at the lowest credible monthly price in the category than charge a big upfront fee for a half-baked promise.

Yes — up to 3 devices on one Pro subscription. Your work laptop, home Mac, spare Windows box. No per-seat fees.

You drop back to the free tier on the next billing date. The bigger models become unavailable, and dictionary/snippet auto-replace stops running (they're Pro features). Your local data — history, dictionary entries, snippets — stays in place. Your account stays. You can resubscribe any time.

30 days, no questions. Email [email protected] and we refund. We'd rather you not pay than feel stuck.

Not yet. The free tier is meant to cover students — it's truly unlimited on the compact models, which is enough for note-taking and essay drafts. If you specifically need the bigger Pro models for research work, email us — we'll figure something out case-by-case.

Not yet. We're a B2C product first. If you have a team that wants SnailText, email us — we can sort out a per-user invoice if you commit to several seats. A proper Team plan is on the roadmap, no date.

Two common causes:

1. 1. macOS Accessibility permission: you need to grant SnailText accessibility access (System Settings → Privacy & Security → Accessibility). The onboarding tour walks you through this; if you skipped it, re-toggle the permission in Settings.
2. 2. Hotkey conflict: another app might own Ctrl+Space (Win) / Option+Space (Mac). Change SnailText's hotkey in Settings → Hotkeys.

Three things help:

1. 1. Custom dictionary (Pro) — add words SnailText keeps misrecognising. They stop being mistakes after one entry.
2. 2. The bigger Pro models — noticeably better on accents and rare terms, especially in non-English.
3. 3. The Pro alternative engine — for European languages, often more accurate than Whisper's largest model and faster.

Mixed. Local input methods (clipboard paste) generally work; global hotkeys can be flaky depending on the host's hotkey policy. Microphone passthrough also depends on the VM. If it doesn't work for you, we'll refund — email [email protected].

Yes. SnailText uses your system's default microphone, but you can pick a specific device in Settings → Audio. The level meter there shows the live signal so you can confirm before recording.

For text input — usually no, because most games don't accept paste from background apps. For chat overlays (Discord, Steam friends) — yes, those work like any other text field.

A loaded model lives in memory — the bigger the model, the more RAM. The largest Pro model uses about 3 GB. The model auto-unloads after 5 minutes of idle to free up RAM, then reloads on next use.

Email [email protected] with what you saw, what you expected, and the app version (Settings → About). If you can attach the log file (Settings → About → Open log folder), we'll fix it faster.