Use a model on this computer

What this does

You can run Maya's features on a model that runs on this computer. Nothing leaves the computer, there is no message limit, and you do not need the Maya add-on.

The app works with two model servers: Ollama and LM Studio. Install one, start it and download a model in it first.

Set it up

1. Start Ollama or LM Studio. In LM Studio, start its local server.
2. Open Settings › AI › Model.
3. Choose A model on this computer. The app shows "Looking for Ollama and LM Studio on this computer…" and then lists the models it finds under each server's name.
4. Pick a model. Each one shows its size and its grade, for example "8B parameters · Capable".
5. Press Test connection if you want to check it answers.
6. Press Use for this company. If AI is already on, read the data statement, tick the box and press Turn on.

If nothing is found, the app says "No model server found. Install Ollama or LM Studio, start it, load a model, then look again." Start the server and press Look again.

Light and capable models

The app grades each model by its size:

• Light: under 7B parameters, or the size is not known. Light models run the assistant's search, the column mappings, the pay sentence and the ROE code suggestion.
• Capable: 7B parameters or more. Capable models run every feature, including letters, document reading and explanations.

What this model can do lists each feature. A feature that needs more shows "Does not run on this model" and "needs a capable model (7B or larger)". On that feature's screen, the button is replaced by "The model chosen for this company cannot do this."

Good to know

• The app asks only localhost on the two servers' usual ports, and only when Settings › AI › Model is open.
• Models on this computer are not available in the browser edition.
• Small models make more mistakes. Every suggestion still waits for a person to accept it.
• Page images go only to models that read images. Otherwise, forms and letters are read from their text.