AI providers and what leaves your device
AI deck building runs on your device unless you change it. On a computer you can send builds to an OpenAI-compatible server instead. This page covers what each choice sends and how to switch between them.
See which provider a build uses
Section titled “See which provider a build uses”- Open Settings and find AI Configuration.
- The Builds run on line at the top names the model the next build will use.
- Tap AI Provider to choose a different one.
A change applies to the next build you start. If the provider you picked cannot be used, the section says so under Builds run on, and builds use the on-device model instead.
On your device
Section titled “On your device”Auto (default) and Gemma 4 E2B IT (Local) run Google’s Gemma model on your device. Nothing about your request, your deck, or your collection leaves it. The one download involved is the model itself, described on The AI model download.
To go back to on-device builds from another provider, tap AI Provider and choose Auto (default).
Your own OpenAI-compatible endpoint
Section titled “Your own OpenAI-compatible endpoint”On Windows and Linux you can run builds on a server you choose, such as Ollama, LM Studio, or vLLM. This option is not listed on Android.
- Open Settings → AI Configuration and tap AI Provider.
- Choose OpenAI-compatible endpoint.
- Tap the Endpoint row that appears.
- Enter the Base URL, for example
http://localhost:11434/v1for Ollama on the same computer. Fill in API key only if the server needs one. - Tap Test connection. If the server answers, the models it offers appear under Model. Pick one, or type its name. The model has to support tool calling.
- Tap SAVE.
During a build the app sends that server your build request, the deck as it takes shape, and
details of the cards it looks up for the model: name, mana cost, type line, rules text, and
whether you own the card. Your API key goes with every request, including Test connection.
Whoever runs the server can read all of it, and an http:// address on another machine sends
it unencrypted.
The URL, model, and key are stored with the app’s settings on this device. They are not in backups, and Delete All App Data erases them. See Backup, restore, and deleting your data.
Locked options
Section titled “Locked options”The AI Provider list also shows options with a lock icon. They are not available in the app, so they cannot be chosen and no build runs on them.
The full account
Section titled “The full account”The Privacy Policy is the authoritative statement of what leaves your device for every AI provider, including cloud providers. The terms that come with the Gemma model are on the Attributions page.