Installing LymeScribe Server
An overview of the two install paths and what each one puts on the machine. For the click-by-click walkthrough — including purchase, license activation, and connecting your first client — follow the server setup walkthrough; this chapter is the map of it.
Last updated 2026-08-311Do these in order
The single most common cause of a server that installs fine and then won't transcribe is skipping the first step. The installer sets up the app; it does not set up the machine-learning environment underneath it.
- Prerequisites — WhisperX in a Python environment, plus a HuggingFace token. Links below.
- Install the server — the MSI or the pkg.
- First-run setup wizard — in a browser on the server machine, at the admin dashboard.
- Activate your license (or run out the 14-day trial first — it's the same install either way).
2Prerequisites
Follow the guide for your platform before you run the installer:
- Windows + WSL + NVIDIA GPU setup — WSL2, CUDA/cuDNN, and a WhisperX Python virtual environment inside the WSL distribution.
- macOS (Apple Silicon) setup — Homebrew and a native WhisperX Python venv; the server runs it through a normal shell rather than WSL.
- HuggingFace account & token — required on both platforms. The speaker-label models are gated: you accept their license once on the HuggingFace website, then generate an access token.
3Windows — the MSI
Installer file: LymeScribe-x.y.z.msi. This is the server installer, not the free client.
Because the installer is newly signed, Windows SmartScreen may warn you before it has built up download reputation. That's expected — the walkthrough has the exact "More info → Run anyway" steps. What the MSI does:
| What | Where |
|---|---|
| Program files | C:\Program Files\LymeScribe\ |
| Service | A Windows Service named LymeScribe (display name "LymeScribe Transcription Server"), set to start automatically |
| Firewall | An inbound rule for port 58994, scoped to the local subnet |
| Configuration | C:\Program Files\LymeScribe\appsettings.json |
Because the service starts on boot, the machine can be locked or logged out and the server keeps serving.
4Apple Silicon — the pkg
Installer file: LymeScribe-x.y.z-macos-arm64.pkg. It is signed and notarized, so macOS shouldn't warn you at all.
| What | Where |
|---|---|
| Program files | /usr/local/bin/lymescribe/ |
| Service | A per-user launchd agent, com.lymescribe.api, loaded at login and kept running |
| Configuration | /usr/local/bin/lymescribe/appsettings.json |
| Logs | /Users/Shared/LymeScribe/logs/ |
Which platform to choose
Windows + NVIDIA GPU
- GPU-accelerated — fast enough for live dictation at office scale
- Needs WSL2 + CUDA/cuDNN
- Runs as a Windows Service
The right answer if several people will dictate live through the day.
Apple Silicon Mac
- CPU-oriented — batch turnaround rather than real-time
- Needs Homebrew + a Python venv
- Runs as a launchd agent
A reasonable choice for a small practice with a spare Mac that's mostly transcribing recordings.
Either way it's the same server software, the same admin dashboard, and the same port — 58994.
First run — the setup wizard
However you installed, open http://localhost:58994/admin in a browser on the server machine. The wizard walks through, in order: confirming the prerequisites, the WhisperX execution mode and paths, model and device selection, an environment validation check (Python venv, WhisperX, GPU, ffmpeg), your HuggingFace token, and finally an admin password plus your first API key.
The environment validation step is the useful one: if the WhisperX venv is incomplete or the GPU isn't visible, it tells you which specific check failed rather than letting you discover it on your first transcription.
Verifying it works
From any machine on the network, the health endpoint answers without any credential:
curl http://192.168.1.100:58994/api/healthYou should get back a small JSON object reporting "status": "online" along with the configured model and device. If that responds but transcription later fails, the problem is the WhisperX environment, not the network — the health endpoint reports the configured device, not a live probe of it. Add ?deep to the same URL to run the real diagnostics (venv, WhisperX import, GPU, ffmpeg, disk); those shell out and take a few seconds.
Next
- Configuration —
appsettings.json, API keys, and the HuggingFace token. - Administration — connecting clients, the queue, updates, and troubleshooting.
- The full setup walkthrough — the step-by-step version of this page, including licensing.