Settings, pane by pane

Everything configurable lives in one window, reached from the gear in the toolbar (or Ctrl+, / ⌘,). A left-hand rail lists the panes; this page walks them in the order they appear.

Last updated 2026-08-31

Servers Dictation Hotkeys Output Corrections Hosting Models Engine AI License General

Three of those rows come and go. Models and Engine appear once you've turned hosting on or been through hosting setup — before that they'd be panes about a machine that isn't hosting. AI appears when the experimental AI features are enabled. Everything else is always there.

Servers

Where transcription work goes. Add a server by address and access key, and LymeScribe keeps the list — a colleague's hosting machine, a practice server, your own host.

Server list: Add, edit, and remove the servers this client knows about, and see which are reachable.
Task routing: Dictation and file transcription can go to different servers. Live dictation wants whatever is fastest and closest; a long batch file can go to the beefier machine. Set each independently here.

If you're hosting on this machine, your own host appears as a server like any other. Setting up a server for the first time? See the server guide.

Dictation

How speaking turns into text in whatever app you're using.

Microphone: Which input device to record from.
Minimum recording length: Ignore accidental taps of the hotkey — anything shorter than this is discarded rather than transcribed as a stray word.
Text delivery: How the finished text reaches the app you're typing into: simulated keystrokes, or the clipboard. Keystrokes feel most natural; the clipboard is more reliable in apps that fight synthetic input.
Per-app overrides: Pin a specific delivery method for a specific application, for the one or two that misbehave with your default.
On-screen indicator: The recording overlay's appearance while you're speaking.
Translation: Optionally have dictation come out in another language, using a configured AI engine.

On Linux, text delivery depends on your desktop. An X11 session types into other apps exactly as on Mac and Windows. A Wayland session needs a one-time permission — Settings shows an Allow LymeScribe to type button — and until it's granted (or on desktops that don't offer it at all) dictated text goes to the clipboard instead, and says so. LymeScribe tells you plainly which case you're in rather than silently typing nothing.

Hotkeys

The keyboard shortcuts that start and stop recording, plus the permissions those shortcuts need.

Push-to-talk: Hold to record, release to transcribe. The default way to dictate.
Toggle recording: Press once to start, again to stop — better for long passages than holding a key.
Re-insert last transcription: Drop the previous result in again, for when it landed in the wrong window.
Permissions: macOS requires Accessibility and Automation permission before the app can watch for a global hotkey and type for you. This pane shows whether each is granted and links straight to the right System Settings page.

On Linux, global shortcuts under Wayland are registered with the desktop rather than owned by the app, so the desktop has the final say over the combination — see the dictation chapter.

Output

The shape of the text LymeScribe produces.

Default format: Plain prose, or with speaker labels where the transcript supports them.
Batch output: Where transcribed files are written and how they're organized into folders.

Corrections

A find-and-replace dictionary applied to every transcript after it's produced. This is where you teach LymeScribe your vocabulary: a client's surname it keeps mangling, a drug name, an in-house acronym, your own preferred capitalization.

Corrections are organized into groups you can switch on and off, so a set of medical terms and a set of legal ones don't have to live in one undifferentiated pile. They're applied on this machine, to the finished text — the transcription engine itself is untouched.

Hosting

Paid unlock

Turn this machine into a transcription server. Covered in full in Hosting on your own machine; the pane itself holds:

Enable hosting: Starts the embedded engine. Requires an active license or trial.
Let someone connect: The address and access key another device needs to attach to this host.
Model: Which speech model this host is using, with a jump into the Models pane to change it.
Speaker labels: Local speaker labeling for file transcription. Batch files only, never live dictation.

Models

Hosting only

Download, choose, and remove the speech models your host runs. LymeScribe recommends one for your hardware and shows the size of each before you commit to the download. Models range from a fast English-only one at around 78 MB up to the most accurate at about 3 GB; the hosting chapter has the full table and the guidance on which to pick.

On a machine with no graphics acceleration, the largest multilingual models are hidden rather than offered — they'd be impractically slow.

Engine

Hosting only

The engine process itself: which build is installed (CPU, or an accelerated one downloaded for your hardware), its current state, and its lifecycle.

Model lifecycle: Unload after idle frees the memory a loaded model is holding once you've stopped working, at the cost of a short reload next time. Leave it off on a machine with memory to spare and hosting stays instantly responsive.

AI

Experimental

Engine profiles for the Summarize, Meeting Minutes, and Translate actions. Each profile names an OpenAI-compatible endpoint — a local server such as LM Studio, or a cloud provider — along with the model to request and a context budget. API keys are held in your operating system's secure credential store.

Set a default engine here; the rest is in the AI tools chapter and the setup walkthrough.

License

Start the free trial, enter or remove a license key, and see what your license entitles you to — including how many devices may attach to your host at once. The client is free forever and needs nothing here; this pane is about the hosting unlock. Pricing lives on the pricing page, and choosing a tier explains the difference between hosting and a dedicated server.

General

The catch-all pane.

Startup: Whether LymeScribe launches when you log in.
Status bar: What the strip along the bottom of the main window shows.
Setup: Re-run the first-launch setup wizard.
History: Keep history for — Forever, 1 week, 2 weeks, 1 month (the default), 3 months, or 1 year. Older transcripts are cleaned up at launch. This is the setting the archive links back to.
Automatic updates: Whether to check for new versions on launch, plus a manual check. The update feed address is shown for reference.
Diagnostics: Open the activity log and its folder — the first thing to look at when something didn't behave, and the thing to attach if you write to support.
About: Version number and build.

One product, three desktops

The Mac, Windows, and Linux apps deliberately share this layout — same panes, same order, same wording — so a setting you found on one machine is where you expect it on the next. Where a platform genuinely can't do something, LymeScribe says so in the pane rather than hiding it; the Linux Wayland text-delivery note above is the clearest example. Linux users can see everything degraded on their particular desktop listed in one place in the app.