Getting Started

How your agent works

Your agent is an AI assistant that answers your visitors using the knowledge you give it, in the tone you set, and can take a few real actions on your behalf. This page explains the whole mental model in plain language, so you know what to expect before you build.

You do not need to understand the technology to run a great agent. But it helps to know the four parts that shape every answer your visitors get. Think of them as four dials you control:

Knowledge. What the agent knows. It answers from your website, documents, FAQs, and a short description of your business.
Behavior. How it sounds and the rules it follows, written by you in plain language.
Skills and apps. The actions it can take, such as capturing a lead, handing off to a human, or booking through a connected app.
Channels. Where it runs, such as your website widget or a shareable chat link.

The Ultimo Bots dashboard overview, showing an agent with its knowledge, behavior, and deployment in one place — Everything that shapes an answer lives in one dashboard: knowledge, behavior, skills, and the channels your agent runs on.

The one rule that matters most. Your agent answers only from the knowledge you give it. When the answer is not in that knowledge, it tells the visitor it does not have that information instead of making something up. More on this below.

1. Knowledge: what it knows

Everything your agent can talk about comes from the knowledge you add. Behind the scenes this content is read, broken into small passages, and stored so the agent can find the most relevant pieces for any question. You manage all of it on the Knowledge page. There are a few kinds of knowledge, and they all work together:

Your website. Add a URL and the agent reads the pages it can reach. You can refresh it later so the agent stays current. See website links.
Documents. Upload files such as PDFs and Word documents, for example handbooks, price lists, or product sheets. See documents.
FAQs. Hand-write question and answer pairs for the things you want answered a specific way. See FAQs.
Business context. A short description of your company that gives the agent background for every conversation. See business context.

You can also connect cloud sources such as Google Drive, OneDrive, and Notion. No matter where a piece of knowledge comes from, it all lands in the same place and the agent searches across it as one library.

It looks things up when they are relevant

Your agent does not memorise your content word for word, and it does not read your entire library on every message. Instead, for each question it searches your knowledge for the passages that actually relate to what was asked, then answers from what it finds. Ask about returns and it pulls your returns policy. Ask about pricing and it pulls your pricing. This is why adding clear, well-organised knowledge has the single biggest effect on answer quality.

If an answer is wrong or missing, start with knowledge. Most weak answers trace back to content that is not in the knowledge base, or is buried in a page the agent could not read well. Add an FAQ or a clearer document and the next answer improves.

2. Behavior: tone and rules you set

Knowledge decides what your agent can say. Behavior decides how it says it. On the Behavior page you write rules in plain English, and the platform folds each one into how the agent talks. You do not write code or pick from rigid settings, you describe what you want. For example:

End every conversation by asking if there is anything else you can help with.
When a visitor asks about pricing, suggest booking a free consultation call.
Never mention competitor products, even if a visitor asks about them.

Each rule is saved to your agent once, so it costs nothing per conversation and applies from the next message onward. You can add, edit, and remove rules at any time, and the number of rules you can keep depends on your plan.

Behavior rules shape tone and wording. They do not, on their own, take actions such as booking a meeting or looking something up in another system. For actions, you use skills and apps, covered next.

3. Skills and apps: actions it can take

Out of the box, your agent answers questions. With skills you give it the ability to do things as well. A skill is a capability you define on the Skills page: you give it a name, describe when it should be used, and add the tools it may use. The agent then decides, in the flow of a real conversation, when that skill applies and what to ask for. Common examples:

Capture a lead. Collect a name and email or other details when a visitor shows interest, so you can follow up. See leads.
Hand off to a human. When a visitor asks for a person, or the answer is not in your knowledge, the agent can offer to connect them to your team through live chat.
Take an action in a connected app. Book a meeting, add a newsletter subscriber, create a support ticket, and more, using an app you connect once.

You connect apps once on the Apps page, and then any skill can use them. Today you can connect Calendly, Cal.com, Mailchimp, HubSpot, Stripe, and Zendesk. Each action runs under your own connected account.

On Wix, your site is already linked. If your agent is installed from the Wix App Market, it is connected to your Wix site automatically, with nothing to set up on the Apps page. That lets the agent work with your site's own products, bookings, and orders.

Private data is protected. When an action touches a specific person's private records, such as their invoices or their bookings, the agent first asks the visitor to confirm their email with a one-time code. The action is then tied to that verified email, so a visitor can only ever act on their own data. This is built in, and you do not have to set it up.

4. Channels: where it runs

One agent can answer in several places at once. You choose where it appears on the Channels page, and you control its look and welcome message on the Design page. The same knowledge, behavior, and skills apply everywhere it runs:

Website widget. A chat bubble you add to your own site. This is how most visitors will meet your agent.
Shareable chat link. A hosted page you can send in an email or post anywhere, with no website required.
Social and messaging channels. Connect channels such as Facebook Messenger, Instagram, Telegram, and Slack so visitors can reach the agent where they already are.

How a single answer comes together

When a visitor sends a message, your agent works through it step by step rather than guessing in one shot. It reads the conversation, searches your knowledge for what is relevant, and looks again or takes an action if the first look is not enough. Here is the pipeline behind one reply:

It reads the question in context
The agent considers the visitor's message together with the conversation so far, so a follow-up like do you have it in red? is understood against what was already discussed.
It searches your knowledge
Before answering, it looks through your knowledge for the passages most relevant to the question. It can search more than once, and refine what it looks for, if the first results are not quite right.
It decides what to do next
With the relevant content in front of it, the agent either answers, looks deeper, or uses a skill, such as capturing a lead or booking through a connected app. It keeps going until it has what it needs.
It writes the reply, grounded in what it found
The final answer is written from the knowledge it retrieved and follows your behavior rules. It replies in the visitor's language. If the information simply is not there, it says so rather than inventing an answer.

Why this matters. Because the agent always looks at your actual knowledge before it answers, its replies stay tied to your real content. It is not reciting general facts from the internet, it is using what you gave it.

What an agent does not do

Being clear about the edges keeps expectations honest and helps you set your agent up well:

It does not invent answers. If something is not in your knowledge, it will say it does not know rather than guess. The fix is to add the missing knowledge, not to expect the agent to fill the gap on its own.
It does not learn on its own from chats. Your content is not silently absorbed into the model. You stay in control of what it knows by editing your knowledge and FAQs, and you can promote a great answer from a real conversation into a saved FAQ.
Behavior rules are not actions. A rule changes tone and wording. To book, look something up, or send something, you add a skill with the right tool.
It only takes actions you set up. The agent can use the skills and connected apps you give it, and nothing more. App actions run under your own connected account.

Always test before you publish. Knowledge, behavior, and skills work together in ways that are easier to feel than to predict. Chat with your agent first, ask the questions your visitors will ask, and refine. See testing your agent.

Put it together

Now that you have the mental model, set each dial in turn. A good order is knowledge first, then behavior, then any skills, then choose where it runs.

Add knowledgeGive your agent something to talk about: website, documents, and FAQs.Set its behaviorShape tone and rules in plain language, no code required.Give it skillsLet it capture leads, hand off to a human, or act in a connected app.Choose where it runsAdd it to your website, share a chat link, or connect a channel.