Fully local speech-to-text.
Cloud-quality transcription and editing.

Almost entirely. evoglyph makes four types of network calls: (1) model downloads over HTTPS from HuggingFace on first launch only — after that, every dictation is fully offline; (2) license activation and periodic re-validation with LemonSqueezy; (3) Sparkle update checks against evoglyph.com; (4) crash reports to Sentry, but only if you explicitly opt in via Settings → Diagnostics. Your audio never leaves your device. See the full network event breakdown.

We recommend against it. evoglyph automatically skips native macOS secure fields — when the focused element is a native password box, dictation is refused and nothing is typed. But in web browsers and some apps, a password field can't always be detected, so evoglyph can't guarantee it will skip them; treat browser password fields as unprotected. Dictation also runs through speech recognition and an AI cleanup pass, both non-deterministic — the inserted text may not exactly match what you said. For both reasons, type passwords and other secrets manually rather than dictating them. See how text injection works.

Parakeet TDT on the Apple Neural Engine matches cloud-grade quality on developer dictation. Latency is typically well under 500ms for short sentences. We ship parakeet-tdt-v2 by default.

The app installer is about 150MB. On first launch, evoglyph downloads the local transcription model (~2.6GB) and cleanup model (~1GB) in the background. Once they're on disk, every dictation runs locally - no further downloads, no server round-trips.

Any Mac app. evoglyph uses three injection methods: Accessibility API splice for native apps, keystroke injection for terminals, and pasteboard for browsers and Electron apps. We have verified compatibility with iTerm2, Obsidian, Chrome, Firefox, Cursor, VS Code, Slack, Linear, Notion, and all standard Apple apps. If an app accepts keyboard input, evoglyph works in it.

Your text lands in the app where you started recording, not wherever focus is when you stop. evoglyph captures the target at the moment the hotkey goes down: the AX-splice path writes to the element captured at recording start; the pasteboard path posts Cmd+V to that app’s process ID. Switching focus mid-dictation does not redirect your words. If injection fails entirely, evoglyph copies the text to your clipboard and shows a notification. See how text injection works for more detail.

The trial allows up to 3,000 transcribed words within 7 days of first launch, whichever expires first — no account required. Wave-1 beta participants can redeem a promo code for discounted pricing at the redemption walkthrough.

Each license covers two active devices. To use evoglyph on a third Mac, deactivate one of your existing devices from Settings → License → Devices. Self-deactivation works remotely — the old device does not need to be online. Full details at docs/licensing.html.

You can cancel a subscription at any time from the LemonSqueezy customer portal; access continues until the end of the current billing period. For refunds, email support@evoglyph.com within 30 days of purchase. Full details at docs/licensing.html.

Yes. After the initial model download, every dictation runs fully offline. The only online dependencies are the one-time model download, periodic license re-validation, and the optional crash-report opt-in. See the full breakdown.

Your audio and transcribed text never leave your device. The four network calls evoglyph makes are: model downloads (HuggingFace, first launch only), license validation (LemonSqueezy), Sparkle update checks (evoglyph.com), and opt-in crash reports (Sentry). See the event-by-event breakdown.

The evoglyph roadmap lists what’s in progress, planned, and under investigation at docs/roadmap.html. Entries are issue-cited; no dates are promised.

No — evoglyph is proprietary, commercial software. The privacy guarantee comes from the local-first architecture, not source availability. Audio processing and AI cleanup run on-device with no cloud round-trip. See where your audio goes.

English. evoglyph is validated end-to-end — transcription plus AI cleanup — on English. The underlying speech engine ships a 25-European-language variant (Parakeet TDT v3), but the full pipeline has not been validated on those languages, so we do not claim non-English support. Non-English support will be advertised if and when our eval harness publishes non-English results.

It's posted publicly to LinkedIn and Instagram on launch day. The same code works for everyone in the Wave-1 cohort - copy it from either post, paste it at checkout. See the redemption walkthrough for the full flow.

$1/mo, $5/yr, or $10 lifetime with the Wave-1 code, vs the standard $2/mo, $10/yr, $20 lifetime. Same product, same features, just the cohort price. Walkthrough here.

Pricing returns to standard for new customers. Existing beta licenses remain at the discounted price for the full term of the subscription, or forever for lifetime. The Wave-1 rate is tied to your continuously active license - keep it active, keep the rate. More on the cutoff and Founders perks.