Loomwright
- requiredThe Aardvark Studio server
- requiredAn Anthropic API key
- optionalA text-to-speech endpoint
- optionalA GPU
Requirements
This is on-premise software. The desktop app is the front of it; behind it is a server you deploy and, for some products, a GPU host you already own or rent. For most people reading this, the administrator is also the buyer — so this page is written for one person doing all of it.
One — the desktop Core
The Core is platform-and-architecture bound, so there are four installers and never one Download button. Add-on products are not downloaded at all — they arrive at sign-in, verified against a key compiled into the Core.
| Platform | Artifact | Status today |
|---|---|---|
| Linux (x64) | AardvarkStudio-Core-1.0.0-linux-x64.AppImage | available Built and verified. Runs as an AppImage; no install step. |
| macOS (Apple silicon) | AardvarkStudio-Core-1.0.0-macos-arm64.dmg | unsigned The build is written but has never been run, and the Developer ID certificate has not been requested. It is not offered here until it opens on someone else’s Mac without a Gatekeeper warning. |
| macOS (Intel) | AardvarkStudio-Core-1.0.0-macos-x86_64.dmg | unsigned Same as Apple silicon: written, unsigned, not offered. |
| Windows (x64) | AardvarkStudio-Core-1.0.0-windows-x64.exe | blocked Blocked on a build host and an Authenticode certificate. There is no Windows build to download and we are not going to imply there is. |
Only Linux is genuinely shippable today
The macOS build exists in the release process but has never been run and its Developer ID certificate has not been requested; the Windows build is blocked on a build host and an Authenticode certificate. Both are procurement, not engineering — and neither is offered here until it opens on somebody else's machine without a warning.
Two — the server
The server owns everything: the books, the audio, the runs, the reviews, the renders and the exports. It ships as a container image plus a compose bundle, versioned in lockstep with the desktop Core, and it is never inside a client installer.
The shape of it
# load the image you were given
docker load -i AardvarkStudio-Server-<v>-linux-amd64.tar
# unpack the compose bundle beside it
tar xzf AardvarkStudio-ServerCompose-<v>.tar.gz
# configure: one secret, one API key
cp compose.env.example .env
$EDITOR .env
# up
docker compose up -dIllustrative. The bundle ships its own README, and that is the one to follow.
Three — the services behind it
An unset service URL means "not deployed", not "broken". The server ships a script that reports exactly which of these this install needs, given what you are licensed for.
| Service | Need | Used by | What for |
|---|---|---|---|
| AI provider (Anthropic API key) | Required | Everything | Every assistant, analysis and script-generation stage calls it. Without it the server starts and those stages fail one by one. |
| Text-to-speech endpoint | Conditional | Core narration · LitCaster | Synthesised narration and every LitCaster speech segment. Not needed if your audio is recorded. |
| ComfyUI render host (GPU) | Conditional | VisualBook · LookBook | Casting images, storyboard frames, assembly, style previews — and the host whose model directory a LookBook publish reads. |
| LoRA training host (GPU) | Conditional | VisualBook · LookBook | Identity training for casting, and style training for LookBook. |
| Transcription service | Conditional | Core ingest | Transcribing the audiobook during ingest. |
| Captioning service | Optional | LookBook | Replaces caption stubs with what each image shows. Manual otherwise. |
The GPU is yours
There is no hosted rendering tier and no bundled cloud GPU. VisualBook and LookBook call a render host you point them at — a workstation on the same network is fine. If you do not have one and do not want one, the Core, Loomwright and LitCaster are the parts of this product that will work for you.
By product
Doing it
None of this needs an IT department. It needs a machine that can run a container and stay on.
Pick the host
Anything that runs a container runtime and has room for your books. A NAS, a spare desktop, a small server.
Deploy the server
Load the image, unpack the compose bundle, set a session secret and your AI provider key, bring it up.
Point it at your services
Only the ones your products need. Leave the rest unset — that is a supported state.
Install the desktop Core
On each machine that will use it. Sign in; licensed products install themselves.
Add a book
A portfolio, then an ePub and its audiobook. The ingest does the rest.
Data
To your AI provider
Book text, prompts and the context each stage assembles. This is the one outbound path that is not optional if you want the assistant and the analysis stages.
To your own services
Whatever you send your render, training, transcription and speech hosts — machines you chose and control. The desktop app's performance telemetry goes here too: it is written locally and shipped to your own Aardvark Studio server, and it stops there.
To us
Nothing. The app does record performance telemetry, and it reaches your server and no further — there is no path from the desktop app or the server to any infrastructure of ours, and an air-gapped install is a supported configuration.
The plain-language version, and the legal one, live on thelegal page.