AI Assistant: chatting with Univents

The Univents Assistant is a chat window built right into your account. It knows your events, your team, and your inquiries, answers questions in plain language, and can take care of tasks for you when you ask. That saves you from clicking through lots of menus.

Opening the assistant

The assistant lives as a round icon in the bottom-right corner of your screen. Here's how to start a conversation:

1. Click the round chat icon in the bottom right (Open Univents Assistant).
2. The chat window opens. The first time you open it, you'll see a short greeting and a few example questions.
3. Type your question into the Ask me anything… (Enter to send) input field and press Enter to send it.

To close it, click Close (the X) at the top right of the window, or press the Esc key. Note: the assistant is currently only available on desktop, not on mobile. The window stays open while you work, so you can ask a new question at any time.

Asking questions and using examples

You don't need to memorize any commands, just write the way you'd speak. When the conversation is still empty, the assistant greets you and shows three clickable suggestions under For example:

• Which events are today?
• Create a quote with the Wedding template
• Who's available next week?

Clicking one of these suggestions drops the text straight into the input field, where you can tweak it or send it right away. The assistant also takes into account which page you're currently on: the current section is shown at the top of the window. So if you ask a question while you have an event open, the assistant usually knows which event you mean without you having to name it again.

What the assistant can do

The three examples above are just a taste. The assistant covers virtually every area of Univents — here's an overview of what you can use it for:

The assistant also covers further areas, such as locations, booking pages, automation rules, and your workspace settings.

Finding customers by product (reverse search)

You remember a dish but not who ordered it? Just ask the assistant:

• "Who has ordered the meatloaf platter before?"
• "Which customers had breaded schnitzel on a quote?"
• "Show me every customer who booked coffee catering, sorted by revenue."

The assistant searches all quotes, invoices and delivery notes in your workspace — including line items you typed in freehand and never linked to a catalogue product. You get back the customers with the number of documents (quotes and invoices counted separately), the quantity ordered, net revenue, and the date of the most recent document. Handy for targeted repeat offers: "Who had this last year but not yet this year?"

Two things worth knowing:

• Older wordings are found too. If a product was renamed in the meantime, the search still finds the older documents, because it also matches through the product in your catalogue.
• You need access to finance documents. The answer comes from your quotes and invoices. Without that permission the assistant tells you so instead of showing an empty list.

Editing, reordering and deleting line items on a document

On a draft, the assistant can not only add line items but also change and reorder them:

• Change: “Set the quantity of Buffet Standard to 135”, “The unit price for travel is €45”, “Give that line a 10% discount.” You can change quantity, unit price, label, description, discount, tax rate and units.
• Reorder: “Move travel and beer tables to the end.” The assistant reorders a whole section.
• Delete: “Take the napkins off.”

A few things that save you work:

• The name is enough. You don’t have to look a line item up — just call it what it says on the document. If the name matches several rows (“buffet” with three buffet lines), the assistant asks and shows you the matches to choose from, instead of guessing.
• The document number is enough. “On ANG-0042, change the quantity to …” works directly; you don’t have to open the document first.
• It quotes the new figures. After the change it tells you the new line total and document total — the ones actually saved, not its own arithmetic.

Deleting affects this document only

Ask the assistant to remove a line item and it removes it from this document only — exactly like the X in the line-item editor. The linked booking item and the same line on other drafts of the same event stay where they are.

If you really want all of it gone, say so explicitly (“from the booking too”, “from every document”). The assistant then shows you a confirmation card listing the affected documents first and waits for you to Confirm.

When a change reaches further

Some line items are connected: an item that came from a booking often also appears on a delivery note, and the components of a package are recalculated when you change the package quantity. The assistant tells you when a change reached other documents. If it cannot confirm what happened, it says that too — rather than claiming nothing else was touched.

Sent documents stay locked

On a sent quote, an accepted order or an issued invoice the assistant changes nothing, not even the order of the lines. It explains the route instead: create a new quote, or correct the invoice with a correction or credit note. It only creates a new draft if you ask it to.

One more thing: adding, changing and deleting line items requires permission to edit finances. With read-only access to finances the assistant says so; viewing documents still works.

What the assistant may do depends on your role

The assistant never has more permissions than you do. It is a second way to operate Univents, not a shortcut around your permissions — in the chat and through an external connection (Claude, for example) alike.

Two levels count separately here:

• Viewing and reporting needs view permission for the area in question. Without it the assistant says so openly instead of showing an empty list.
• Creating, changing, deleting or sending something needs the same create, edit or delete permission you would need to click it in the app. If you may only view an area, the assistant tells you the permission is missing — it does not carry the action out on your behalf instead.

This covers, for example, granting or revoking app access, seating plans and function sheets, bulk messages to crew or customers, sending and replying to emails from the mailbox, quotes created from an email, and custom holidays.

If someone should be able to do more through the assistant, you change that person's role under Settings → Roles, not the assistant.

Asking “where do I find …?”

If you are looking for a particular area, just ask — for example “where can I change the line items” or “where do I set the VAT rate”. The assistant gives you the menu path in a single line and links the help article the answer came from underneath. It then offers to take you there; nothing happens until you say yes.

If the documentation does not cover the path, it says so plainly and offers to navigate you to a related area instead. It never guesses — an invented menu path would cost you more time than an honest answer.

Complete lists: the assistant says what it has not seen yet

When you ask about a lot of records — “how many products do I have?”, “list every item under €1” — the assistant loads them page by page rather than all at once. It knows the exact total while doing so, and tells you when it has only seen part of it: it then names how many entries are still missing, or loads them, instead of taking the first page for the whole set.

In practice that means:

• “That’s all of them” only comes once it really has loaded all of them.
• “There is no such product” only comes once it has been through every page — before that, it keeps loading.
• If you want a real total, just ask for it (“how many are there in total?”). The number comes from the database, not from an estimate.

If you filter while doing so (by a category or a search term, say), the total refers to exactly that filter.

What the assistant will not claim

The assistant answers only from what it has actually read — from your data, or from a source it backs up with a number such as [1]. Where it has neither, it says so and offers to look it up, instead of producing a plausible guess. That covers three cases in particular:

• Numbers and quantities. Why a position carries a certain quantity, how many records exist, what something costs — it looks that up.
• “Does feature X exist?” Whether a setting, an automation action or a field exists is something it checks, rather than confirming or denying it from memory.
• Formats and legal matters. Which e-invoicing format is produced, which deadline applies: with a source, or not at all.

An honest “I don’t know — shall I check?” is deliberate here. An invented number costs you more time than a follow-up question, and in case of doubt there is no way to tell that it is wrong.

The same principle applies to deleting: if a product was already deleted, it says “was already deleted” instead of showing an error. If it finds nothing for an id, it does not try the next one — it looks for the right record.

Dictating by voice

If your browser supports it, you can speak your question instead of typing it. A microphone icon then appears next to the input field.

1. Click the microphone (Start voice input).
2. Allow the browser one-time access to your microphone if it asks.
3. Speak your question, the spoken text appears in the input field as you talk.
4. Click the button again (Stop voice input) to end the recording.

Any text you've already typed stays put, the dictation is simply appended to it. You can then correct the text and send it with Enter as usual. If no microphone appears, your browser doesn't support the feature, so just type your question instead.

Attaching images and files

Next to the input field you'll find a paperclip. Use it to attach an image (a screenshot, for example), a PDF, a Word document, or an Excel sheet to your message — the assistant sees the content and can work with it, for instance turning a list in the image into tasks.

Good to know: the assistant keeps the most recently attached images in view — so you can still refer to a screenshot a message later ("now turn that into tasks"). If you attach many images one after another, the three most recent stay visible; older ones it can no longer look at, and it will say so rather than guess — just attach the image once more. For PDFs, Word, and Excel files, the text stays available for the whole conversation.

Choosing a model: Simple or Complex

Above the input field you'll find the Model switcher with two options: Simple and Complex.

• Simple is the default and fast enough for most questions and tasks.
• Complex uses a more powerful model suited to harder tasks — the label right next to the button flags a higher cost per message as a result.

Your choice applies for the current conversation: as long as you stay in the same conversation, the setting sticks, even if you navigate to a different page in the meantime. A new conversation always starts back on Simple. While an answer is streaming, the switcher can't be used.

When your AI credit balance runs out

Both questions and actions from the assistant consume your workspace's AI credits — Complex consumes more than Simple. If the balance is empty, the assistant shows the message AI credit used up. Top up your balance to keep using the assistant. For how to check your balance, top it up, or set up automatic top-ups, see AI Credits & Usage.

Confirming actions

The assistant doesn't just answer questions, it can also get things done, for example sending a document or recording a payment. So that nothing happens unintentionally, it asks before such steps. A card appears with a heading describing what it plans to do, a short summary, and the details.

• Confirm runs the action.
• Cancel discards it, and nothing happens.

After you confirm, the card shrinks down to a short line. Using Details, you can expand it again to see exactly what happened and when. If a confirmation sits too long, it expires (Confirmation expired — please enter again); just enter your instruction again.

Large jobs: the assistant pauses instead of stopping

For jobs that take many steps — “create these 40 items”, say, or “clear out the test products” — the assistant works through them one step at a time. Two limits can come up along the way:

• It has reached the limit of 16 work steps per message.
• It is approaching the time limit for a single message.

In both cases it does not stop silently. It writes a note into the chat and pauses. Just type weiter (“continue”) and it picks up exactly where it left off. It never redoes steps it has already completed.

Everything the assistant did up to the pause stays in the conversation, so you can always see how far it got.

If you reload in the middle of a job

The conversation is saved continuously while the assistant works, not only at the end. If you reload the page in the middle of a long job or close the tab, everything up to that point is still there when you reopen the conversation. An interrupted job no longer leaves you with an empty chat.

One job per conversation

Within a conversation, the assistant only ever works on one thing at a time. If you send a second message while a job is still running — from a second tab, say, or right after Stop answer — it is not accepted, and you get: Another request is still running in this conversation. Please wait until it finishes, then send your message again.

That is deliberate: two simultaneous jobs in the same conversation could carry out the same instruction twice — deleting or creating the same products twice, for example. Your message stays visible in the conversation, so you can copy it from there and send it again once the running job has finished.

If you want to ask something else in parallel, start a new conversation with the pencil icon — you can keep working there right away.

Finding earlier conversations

Your conversations are saved, so you can return to an old thread at any time.

1. Click the speech-bubble icon at the top of the window (Show conversations).
2. The Conversations view shows all your previous threads with title and date.
3. Click an entry to reopen that conversation.

If you'd like a fresh start, click the pencil icon at the top (Start a new conversation). While an answer is in progress, you'll see the Answering… indicator, and the send button turns into a stop button: with Stop answer, you can interrupt an answer in progress at any time.

Tips

• Be specific: The more precise your question, the more precise the answer. Instead of "Show me events," try "Which events are this week?"
• Use the context: Open the relevant event or inquiry first, then the assistant can refer to it directly.
• Check confirmation cards: For actions like sending documents, read the summary briefly before you click Confirm.
• Check the amounts on documents: When the assistant adds line items, it enters the unit price and the quantities separately (e.g. 1 × 3 people × 3 hours × €39.92). For products billed per person or per hour, take a look at the line total before you send: the quantities come from your event, so if the guest count or duration there differs from what you agreed, the line item is calculated with those.
• Follow the sources: For questions about Univents features or your knowledge base, the assistant marks every backed-up statement directly in the text with a small numbered citation like [1]. Tap or hover over it for the text excerpt; for help articles, clicking it opens the article in a new tab. More on this in AI Answers with Source Citations (Knowledge Base).