Where your audio goes

Last updated 4 min read

From hotkey press to text in your app, every stage of dictation runs on your Mac. This page maps the full pipeline, lists every network event the app makes, and shows how to verify all of it yourself.

The dictation pipeline

Nothing in the dictation path touches the network.

Stage What happens Where it runs Leaves device?
Mic capture The app captures microphone audio; it never leaves your Mac. On your Mac No
Speech-to-text FluidAudio's Parakeet TDT model converts audio to text on the Apple Neural Engine. The model runs entirely on-device after the first-run download. Apple Neural Engine (on your Mac) No
AI cleanup When cleanup is enabled, an MLX-powered LFM2-2.6B 4-bit model runs inside the Evoglyph process. The raw transcript and the cleanup prompt stay in process memory; no network calls at inference time. In-process on your Mac No
Text injection Evoglyph inserts the final text into your focused app via macOS Accessibility APIs (AXUIElement splice for native apps, keystroke injection for terminals, pasteboard for browsers and Electron apps). On your Mac No
History database Dictation history lives in a local SQLite database on your disk. Evoglyph has no copy of this data and cannot access it remotely. We recommend FileVault for at-rest encryption. Local disk on your Mac No
Vocabulary and settings Custom vocabulary, hotkey choice, and all other preferences stay in local app storage, never synced to a server. Local app storage on your Mac No

Network events

Evoglyph is local-first, not network-free. This table lists every outbound connection the app makes. None of them touches the dictation pipeline.

Event Endpoint When Payload
Model download Hugging Face model hosts (over TLS) First launch only; fully offline after that Standard HTTPS download request. No audio, no transcripts, no personal identifiers beyond what any HTTPS request carries.
License activation & validation evoglyph.com (licensing service) On license key entry; periodically re-checked to keep subscription state in sync (active, past-due, canceled) License key, anonymized device identifier, Evoglyph version. No dictation content. The licensing service is Evoglyph's own; see Licensing and activation for details.
Free-trial start & usage sync evoglyph.com/api/trial/claim and evoglyph.com/api/trial/sync (licensing service) Trial only: once at first launch (start), then periodically after dictations (usage sync) until the trial ends A pseudonymous device hash and the trial’s cumulative word count — a single number. The response is the signed trial state (start time, word total, limits). No audio, no transcripts, no app names, no name or email. See the privacy policy for what the trial ledger stores.
Update check evoglyph.com/appcast.xml (Sparkle framework) On a schedule; can be disabled in Settings Only your Evoglyph version (in the request user-agent) and your IP address, not your macOS version. Checks are counted in aggregate (date, country, Evoglyph version); no IP address, device identifier, or cookie is stored.
Crash and error reports diag.evoglyph.com (Cloudflare Worker on Eluketronic’s own account; no third-party telemetry vendor) Only when you opt in via Settings → Diagnostics; off by default Unsymbolicated stack traces, Evoglyph version, macOS version, speech-model variant, cleanup-engine state, pipeline state at event time, recent app-event categories (never message text). No audio, no transcripts, no vocabulary, no clipboard, no app names. Anonymous: no account, device, or user identifier. Deleted after 90 days.
In-app analytics None Never Evoglyph does not collect in-app analytics or usage telemetry. The marketing site is measured server-side at Cloudflare’s edge (no analytics script, no cookies); the app sends nothing.

Local vs cloud dictation tools

Most cloud dictation tools stream your microphone audio to a remote server for transcription, then return text over the network. Every spoken word travels off your device, and latency includes a round trip to a data center.

Evoglyph runs the transcription model on the Apple Neural Engine inside your Mac; your audio is never transmitted. Latency is the time the Neural Engine takes to run the model: no network round trip. The tradeoff is a model download on first launch and an Apple Silicon Mac requirement.

Verify it yourself

You do not have to take our word for the pipeline table above. Two independent checks:

For the full privacy policy, including third-party processors, data retention, and your rights, see Privacy and local-first.

Was this helpful?