Downloading and managing local models

Arqenne runs its on-device AI features on your own computer, on macOS and on Linux. To do that it needs a few model files: the AI assistant, the search model and the voice engines. You download each one once, from Settings › Models (or from the prompt on the home screen), and after that it works offline.

This guide explains what each model does, how big it is, where it lives, and how to manage it.

What gets downloaded

Arqenne has four on-device models. Each powers a different feature, and you choose which ones to download:

A few things worth knowing:

• Nothing downloads behind your back. Each model has its own Download button in Settings › Models, and the button shows the size before you start.
• Models load when you first use them. A downloaded model doesn't slow down app startup; Arqenne loads it into memory the first time a feature needs it.
• The Local AI Assistant picks its variant for you. On computers with 12 GB of memory or more it uses a larger, higher-context variant; everything else works the same.
• Enhanced Text-to-Speech is offered only where it can run. On a Free or Starter plan, or on a computer with less than 12 GB of memory, its card explains why instead of offering the download.

All four together take roughly 15 GB on macOS and 13 GB on Linux. If you only narrate with the Standard voices, you need just ~100 MB.

Models download from Arqenne's model storage the first time only. After that, generation happens entirely on your computer — your text and audio never leave it.

Watching the download

Downloads need a network connection. While one is running, its card in Settings › Models shows progress, and you can keep working in the rest of the app. The Local AI Assistant and Enhanced Text-to-Speech are the largest, so on a slow connection they take a while.

Once a model is downloaded, you can disconnect from the network entirely. Everything that model does is local and offline from then on.

Where models are stored

Downloaded models live in Arqenne's app data folder:

Inside that folder, each model has its own subfolder:

You do not need to open these folders for normal use — Arqenne manages them for you. They are listed so you can check disk usage. To remove a model, use Settings › Models (see Reclaiming disk space) rather than deleting folders by hand.

System requirements

• macOS: an Apple Silicon Mac with macOS 13 Ventura or later. The Local AI Assistant and Enhanced Text-to-Speech run on the Mac's GPU (Metal).
• Linux: a 64-bit Intel or AMD PC with Ubuntu 24.04 LTS or newer, or Debian 13. The Standard edition runs every model on the processor; the NVIDIA edition runs the Local AI Assistant and Enhanced Text-to-Speech on an NVIDIA graphics card. See Arqenne on Linux for details.
• Memory: 8 GB covers the Standard voices and the Local AI Assistant. Enhanced Text-to-Speech and voice cloning need at least 12 GB.
• Disk: about 20 GB free covers the app and all of its models.

Generated audio is 24 kHz WAV in every case.

Managing models

If a download is interrupted

If you quit Arqenne, lose your connection, or restart mid-download, nothing is lost. Open Settings › Models and press Download again: files that already finished are kept, and only what is missing is fetched.

If a download seems stuck rather than simply slow, quit and relaunch Arqenne — see Troubleshooting below.

Reclaiming disk space

To remove the Local AI Assistant or either Text-to-Speech model, open Settings › Models, open the ⋮ menu on the model's card and choose Delete. Arqenne frees the space and marks the model as not downloaded. (AI Embedding has no delete option.)

Deleting a model is reversible, but it is not free. To use that feature again you download the model again from Settings › Models — which means a network connection and another download of the same size. Only remove a model if you genuinely need the space.

The Local AI Assistant (~10 GB) and Enhanced Text-to-Speech (~3.5 GB on macOS, ~2.1 GB on Linux) take back the most space. Standard Text-to-Speech frees only ~100 MB.

Troubleshooting

A download is stuck or failed

1. Check your network connection. The first download of each model needs internet access.
2. Quit Arqenne and relaunch it, then press Download again in Settings › Models. Files that already completed are not fetched again.
3. If it still will not progress, confirm you have enough free disk space (see below), then try again.

Not enough free disk space

Make sure you have room for the model you are downloading — its size is on its Download button. Free up some space, then start the download again.

A feature is "loading" the first time you use it

Arqenne loads a model into memory the first time a feature needs it, not at startup. That first use takes a few seconds longer, and narration or ✦ Suggest may show a loading state. This is normal; wait a moment and it becomes ready.

A feature never becomes ready

Check Settings › Models: each card shows whether its model is downloaded, loading, or reporting an error. If it stays in an error state, quit and relaunch Arqenne. If the problem persists, delete the model from its ⋮ menu and download it again.

Privacy

• Models download once, from Arqenne's model storage, when you press Download.
• After that, nothing is sent anywhere. Narration, voice cloning, the Local AI Assistant and writing suggestions are generated on your computer. Your manuscripts, your text, and your audio never leave it.
• You are in control. You choose which models to download, and you can delete them from Settings › Models at any time.

See also

• Narrate your manuscript with Text-to-Speech — the Audio editor, paragraph lanes, and Play All
• Clone a voice with Enhanced Text-to-Speech — create a custom voice from a short recording
• AI writing suggestions — select text and get on-device edits

Need the app? Download Arqenne → · Curious about voices and plans? See pricing →