Setting up Dax

The add-on

The first time you open Dax, Deskboard downloads the Dax add-on, which is around 400 MB. This happens once per computer, not once per board.

Choosing a model

Dax needs a model before it can answer anything. Open Provider settings from the top of the Dax panel. It has three options: API Key and Local Models and MCP Server.

You can always switch between them.

	Runs on	Good for
API key	API provider’s servers	The strongest models, and nothing to download
Local model	Your own computer	Free usage and keeping everything on this machine

Using an API key

On the API Key tab, pick your provider from the list. OpenAI, Claude, Gemini, OpenRouter, Groq, DeepSeek and NVIDIA are all there, and Other supports any OpenAI compatible API endpoint.

Paste your key, then set the Model. You can type the model name if you know it, or click browse to pull the list from the provider.

The free trial

If you do not have a key, there is a free trial option that uses a shared NVIDIA key to get you started.

Caution

The trial key shares rate limits between everyone using it, and NVIDIA may collect the data sent through it. Use your own key or a local model for anything private.

Reasoning

Models that can think before answering take a Reasoning level: None, Minimal, Low, Medium, High or Extra high, with Maximum available on Claude. Leave it on Default unless you have a reason to change it. Higher levels give better answers on hard questions and cost more.

A model that does not reason ignores this setting.

Reply limit

Under Advanced options, Reply limit caps how long a single answer can be. Leave it blank to use the provider’s own default.

Running a model on this machine

Local models run on your local machine. It works offline, remains free and nothing is sent to any server.

Tip

Local models rely on your GPU and RAM. The better your PC, the more powerful models it can support.

Browing Models

Browse models searches Hugging Face. You can also paste a Hugging Face repository address directly if you already know what you want.

Bigger models give better answers and need more memory and disk space. Models that can look at pictures are marked Vision.

Click Download to fetch one. The download shows progress and can be paused, resumed or stopped, and a half-finished download picks up where it left off.

Your models

Your models lists what you have already downloaded. From here you can:

• Use this model to make it the one Dax answers with, shown as In use
• Delete from this machine to free up the space

A model marked Missing was deleted outside Deskboard and needs downloading again.

The MCP Server tab

You can also connect your Deskboard to any other AI app. See Connecting other AI apps.