User guide

Getting started

You have installed XataWorks and opened it. This is what it is, the one thing you have to decide before it will answer, and what to ask it first.

XataWorks is a desktop application with two halves in one window: a chat on the left, and a real web browser on the right. The assistant can see the browser, and — when you let it — work the applications open in it. Your helpdesk, your store, the internal tool your company built.

That is the whole idea, and it is why the rest of this guide spends most of its time on what the assistant may do to a website and what it asks you first.

Install it

Download XataWorks and run the installer for your platform. There is nothing to configure during installation and no account to create.

The first launch

The window opens with an empty chat on the left, the message box at the bottom, and the browser panel on the right. Nothing has been asked yet.

The XataWorks window on opening: a menu bar reading File, Agents, View, Help; a left sidebar with the XataWorks mark, a New chat button, Activity, and one chat called New chat; an empty chat area with a Message box and a Send button at the bottom; and on the right a browser panel with two tabs open on a fictional helpdesk and its billing page. — The application as it opens. The two tabs in the browser panel are a fictional helpdesk used throughout this guide; yours will open on whatever you point it at.

The one thing to decide

XataWorks does not come with a model, and it does not resell one. It runs on an account or a provider of your own, and choosing which is the one decision that has to be made before the assistant will answer anything.

Open File ▸ Preferences… and go to AI Providers. That panel is where the default provider and the model are set, and where a stored key goes if your provider uses one.

The Preferences window. A sidebar lists sections under five headings: Account, with AI Providers; Agents & Skills, with Agents, Skills, Add-ons and Variables; Integrations, with Connections and External connections; Browsing, with Browser, Passwords and WebMCP Log; and Advanced, with Folder Watchers and Developer. The AI Providers panel is shown, headed Assistant (interactive): a Default provider selector reading OpenRouter, a Model selector, a note that a Claude command-line tool was detected on this machine, and a second section headed “API keys for automation (agents & scheduled runs)”. — Preferences, opened on AI Providers. The sidebar is the whole of what is configurable; this guide comes back to most of it. What this panel reports depends on your own machine and your own choices: the figures in this guide were captured with OpenRouter selected, which is not the provider XataWorks ships configured, and on a machine where a command-line assistant was also signed in.

Two kinds of use, two places a credential is read from. An interactive chat can run on a signed-in subscription. An agent working in the background, or a prompt you scheduled, has no window to sign in from, so those use a stored API key instead. The AI Providers panel carries both, and says which is in use.

Ask it something

Open a site in the browser panel, then type into the message box the way you would ask a colleague. Describe the outcome you want rather than naming a command:

Which tickets have been waiting more than a week, and who raised them?
Has this customer’s order shipped yet?
Summarise what is on this board and draft a reply to the oldest three.

The assistant reads the page, uses whatever operations the site offers it, and answers. Each thing it runs appears in the transcript by name, and folds open to show what it was given and what came back.

A question typed into the message box, sent, and answered from the site’s own data — then a follow-up in the same conversation. The steps are: open the site in the browser panel, click into the message box, describe the outcome you want, press Send. This recording has no sound.

Two things that will save you a puzzled minute

The assistant can only act on tabs this chat has open. If you ask about a site and it says it cannot see one, the usual reason is that the tab is open in a different conversation. Ask it to open the site and it will.

It will ask before it changes anything. Reading a page and acting on it are separate permissions, and an operation that changes something is confirmed before it runs. That is not a setup step you have missed; it is the design, and Approvals and permissions is the article about it.

Where to go next

A tour of the window names everything on screen. Where the agent’s abilities come from explains why it can do something on one site and not another, which is the question most people have second.

Back to all user-guide articles.