← App User Guide

Hosting on your own machine

Everything so far assumed something was doing the transcribing. Hosting is how that something becomes your computer — the paid unlock turns LymeScribe from a client into its own transcription server, with the engine running inside the app and nothing leaving the machine.

What the unlock actually turns on

LymeScribe is free forever as a client: install it, point it at a LymeScribe server someone else runs, and you get the full dictation experience. What the one-time Host unlock adds is the other half — your machine becomes the server. LymeScribe starts a transcription engine bundled inside the app, transcribes your dictation and your files locally, and can serve a few other devices on your network at the same time.

There is no cloud step in either mode of hosting. The audio, the model, and the transcript all stay on the computer doing the work.

Turning hosting on

1. Unlock it. Open Settings and go to the License pane to start a trial or enter your license key.
2. Open the Hosting pane. This is where hosting is switched on and where the engine and its options live.
3. Download a model. The first time you host, LymeScribe needs a speech model on disk. The Models pane recommends one for your hardware and downloads it for you — see below.
4. Watch the status strip. Along the bottom of the main window, the engine reports what it's doing: starting, ready, which model is loaded. When it says ready, you're hosting.

Choosing a model

The model is the file that does the actual listening. Bigger models are more accurate and slower; smaller ones are faster and lighter. LymeScribe ships a short, curated list rather than the full zoo, and recommends one based on your machine.

Two things worth knowing. The English models are English-only — if you dictate or transcribe in another language, choose Small or larger. And on a machine with no graphics acceleration available, the big multilingual models are hidden rather than offered, because they'd be impractically slow; the app would rather show you a shorter honest list than let you pick something that takes an hour.

Models are downloaded once and reused. You can switch models later from the Models pane; the engine reloads with the new one.

What runs the engine on each platform

The transcription engine is the same across all three desktop apps, but the acceleration available to it is not. This is the honest version:

You don't pick a variant by hand. LymeScribe looks at the machine, offers the accelerated engine when your hardware supports it, and falls back to CPU when it doesn't. Downloaded engines are kept, so it's a one-time step per machine.

Who can attach, and what happens when two people go at once

Hosting is designed for you and a handful of devices — your laptop, a second machine on your desk, a colleague across the room. The unlock allows roughly three attached peers, and it is important to read that as a connection cap, not a throughput promise:

1. Connections are capped. A limited number of clients may be attached to your host at one time. The serving strip at the top of the window shows your host identity and how many peers are currently attached.
2. Jobs queue; they do not run in parallel. If three people submit work at the same moment, the engine transcribes them one at a time, in order. Nobody's job is dropped, but nobody gets three-way parallel speed either.

For a practice where several people need transcription simultaneously and quickly, that queue is the ceiling you'll hit, and the answer is a dedicated machine running LymeScribe Server rather than a bigger peer count on a desktop.

Reading the serving strip and the status strip

When you're hosting, the main window grows two pieces of machinery reporting.

The serving strip sits at the top and answers one question only: who am I, and who's attached? It carries your host identity and the peer count. Deliberately no engine talk and no meters — that's the other strip's job.

The status strip runs along the bottom and is the single home for every machine fact:

• Engine state and the loaded model — starting, ready, stopped, and which model is in memory.
• Eject / Retry — unload the engine, or restart it after a failure.
• GPU and VRAM meters, shown only while hosting, because only a host has an engine to measure.
• The queue summary — how much work is waiting.
• Activity — opens the log, the place to look when something didn't behave.

If a transcription fails or the engine won't start, the status strip is the first place to look and the Activity log is the second.

Speaker labels while hosting

Hosting includes local speaker labels for file transcription — the step that turns a wall of text into a transcript that says who said what. It runs on batch files only, never on live dictation, and it runs on your machine like everything else. See Naming speakers for the workflow and tuning speaker labels if the results need a nudge.

When hosting misbehaves

• Engine won't start. Check the status strip for the reported state, then Activity for the reason. Retry from the status strip after fixing it.
• No model listed. Nothing has been downloaded yet — open Settings ▸ Models and take the recommendation.
• A peer can't attach. Confirm both machines are on the same network, that the peer is pointed at the right host address, and that you aren't already at the connection cap.
• Transcription is slower than you expected. Either a larger model than the hardware wants, or a CPU-only engine on a machine that has a GPU path available. Both are visible in the status strip.

Still stuck? The FAQ covers the common ones, and support is a real person.