How your agent works
Your agent is an AI assistant that answers your visitors from the knowledge you give it, follows the rules you set, and uses skills to get real things done. This page explains the whole picture in plain language, so you know what to expect before you build.
You build and change your agent in AI Studio, mostly by talking to Mo, the AI builder. You do not need to understand the technology, but it helps to know the parts that shape every answer:
- Knowledge and Business Context. What the agent knows: your website, documents, FAQs, and the key facts about your business.
- Skills and apps. What the agent can do, such as capturing a lead, handing over to a person, or booking through a connected app.
- Behavior. How it talks and what it must never do: your own rules, plus the system instructions every agent starts with.
- Intelligence. How capable its answers are, and how many AI credits each answer costs.
- Channels. Where it runs: your website, a chat link, or messaging apps.
1. Knowledge: what it knows
Everything your agent can say about your business comes from the knowledge you add in the Knowledge Base. The content is read, broken into small passages, and stored so the agent can find the most relevant pieces for any question. The kinds of knowledge work together:
- Your website. Add pages and the agent learns from them. Depending on your plan, they are re-read on a schedule. See Website Links.
- Documents. Upload files such as PDFs and Word documents. See Documents.
- FAQs. Hand-written questions and answers for things you want answered a specific way. See FAQs.
- Connected sources. Google Drive, OneDrive, and Notion.
The Business Context is different: it is a short description of your business that the agent always has in front of it, in every conversation, as verified facts. You also find it in AI Studio under Your agent › Profile.
It looks things up when they are relevant
Your agent does not read your whole library on every message. For each question it searches your knowledge for the passages that relate to what was asked, and if a part of the question is not covered, it searches once more for exactly that part. Ask about returns and it pulls your returns policy; ask about opening hours and it looks for those. Clear, well-organized knowledge has the biggest effect on answer quality.
2. Skills and apps: what it can do
A skill is one thing your agent can do, with its own procedure and, where needed, the tools to act. Every agent starts with Lead Generation, and you can add ready-made skills or have Mo build new ones from your description. Examples:
- Capture a lead. Collect a name and email when a visitor shows interest, so you can follow up. See Leads.
- Act in a connected app. Book a meeting, add a newsletter subscriber, look something up, and more.
Handing over to a person is built in and needs no skill: when a visitor asks for a human and your team is online, the agent can connect them through Live Chat.
For each message, the agent picks the skills that fit. It does not run every skill all the time: a question about prices brings in the skill that knows about prices, a booking request the booking skill. Only the chosen skills' procedures and tools are used for that answer.
Apps are connected once on the Apps tab, and then any skill can use them. Today you can connect apps such as Calendly, Cal.com, Mailchimp, HubSpot, and Stripe. Zendesk is coming soon. Each action runs under your own connected account.
3. Behavior: the rules it follows
Knowledge decides what your agent can say. Behavior decides how it says it. Two layers work together:
- Your behavior rules. Plain-language rules you write yourself or let Mo write for you, for example Always answer formally or Never mention competitor products. Each rule can be switched on or off, and the number of rules depends on your plan. See Behaviors.
- System instructions. A small set of base rules every agent gets from Ultimo, such as staying on topic. You see them in the same place and can switch each one on or off. Your own rules always take precedence.
4. Intelligence and AI credits
In Your agent › Profile you choose how capable the agent's answers are. Ultimo Core is fast and economical, Ultimo Pro is more thorough, and Ultimo Max is the most capable. Each answer costs 1, 4, or 8 AI credits respectively; Pro and Max need a paid plan. The same credits also pay for Mo's work. When an agent's credits are used up, it pauses until you add credits, upgrade, or the next period starts. See AI credits.
5. Channels: where it runs
One agent can answer in several places at once. You choose where it appears in Channels and how it looks in Design. The same knowledge, skills, and rules apply everywhere:
- Website widget. A chat bubble on your own site. This is how most visitors meet your agent.
- Chat link. A hosted chat page you can share by email, on social profiles, or as a QR code, with no website needed.
- Messaging apps. Facebook Messenger, Instagram, WhatsApp, 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, the agent works through it step by step instead of guessing in one go:
It reads the message in context
The agent considers the message together with the conversation so far, so a follow-up like do you have it in red? is understood. It notes the visitor's language and picks the skills that fit this message.
It looks up your knowledge
It searches your knowledge for the passages that matter for the question, and searches again for any part the first results do not cover.
It answers within your rules
With the Business Context, the system instructions, your behavior rules, and the chosen skills in front of it, the agent writes the answer. Where a skill needs it, it uses the skill's tools, for example to check free slots or create a booking, and keeps going until it has what it needs. The reply is in the visitor's language.
The answer can be checked
If Check every answer automatically is on in Profile (it is off by default), each answer is checked: is every statement backed up by your knowledge, and did the actions it mentions really happen? In Test you see the result under the answer, or click Check answer in What happened? to check one answer. For visitors, the check runs right after the answer is sent, so it never slows the conversation down.

What an agent does not do
- It does not invent answers. If something is not in your knowledge, it says it does not know. The fix is to add the missing knowledge.
- It does not learn on its own from chats. You stay in control of what it knows by editing your knowledge, FAQs, and Business Context, and you can turn a good 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, the agent needs a skill.
- It only takes actions you set up. The agent uses the skills and connected apps you give it, and nothing more.
Next steps
A good order: give the agent knowledge, describe your business to Mo, test, then choose where it runs.