Skip to content

How do I ask the assistant something?

The assistant is the ERP’s conversation window. You open it from the floating button, write to it in plain language and it answers in the same window. It is for questions about using the program and for looking up your company’s data.

Where is the assistant?

The assistant lives in the round button that floats over the screen, at the bottom right. You can drag it if it gets in the way.

Clicking it opens a menu with four options: Create, Chat, AI assistant and Setup. The one that opens the assistant is AI assistant.

Watch out for Chat: that is the internal chat between colleagues at the company, not the assistant. They are two different things even though they come from the same button.

The panel opens at the bottom right, with the title Assistant and, below it, whether it is connected and the credits you have left. Until you have typed anything it suggests four questions to start with: click them and it sends them as they are.

How do I ask it a question?

  1. Open the panel from the floating button, AI assistant option.
  2. Type the question at the bottom, where it says Type your question or paste a screenshot with Ctrl+V. The box grows on its own as you type, so you can see all of it before sending.
  3. Click the blue send button, or press Enter. To start a new line without sending it, use Shift + Enter.

The answer writes itself out, word by word. While it lasts, the send button shows three dots and the box is locked: you can’t chain two questions at once.

Answers come formatted: bold, lists and, when there are figures, tables. You don’t need to do anything to see it.

Ask in full sentences. ‘How do I make a credit note for an invoice that has already been issued?’ works better than ‘credit note’. The assistant also understands which screen you mean if you are on it: it gives more weight to the documentation for the module you have open.

Can I see the answer as a report?

Yes. When what the assistant answers is a list of data —customers, invoices, instalments, stock— and the program has a report of that same thing, the 📄 View as report button appears above the typing box, next to ✨ Professional analysis.

They are two different ways of improving what you have just read: the analysis goes deeper; the report gives you all the data and lets you take it away. The assistant only shows a few dozen rows in the conversation; if it tells you ‘you have 3.095 customers, the list is truncated’, the button is the way out.

Clicking it opens the usual report, on its screen, with its preview and its download as PDF or spreadsheet. And it opens with what you had already said already set: if you asked about a customer’s invoices in a quarter, the report comes up with that customer in the range and with those dates. You don’t have to type it again.

If several reports fit, the button drops down up to three for you to choose; under each one you see which filters it will open with. If only one fits, the button takes you straight to it.

The button belongs to that answer: with the next question it disappears. And clicking it doesn’t use credits or count as an assistant query.

It doesn’t always appear. It doesn’t appear when the answer is a single record rather than a list, when the assistant is proposing an action for you to confirm, or when you don’t have permission for any report on that subject. Reports you have created yourself in the designer are also included: it is enough for their title to say what they are about.

Why does it ask me whether to carry on?

Some queries return a lot of data: the balance sheet of a large set of accounts, a general ledger for the whole year, a report with no filters. When what has to be read is a good deal more than usual, the assistant doesn’t answer with half of it: it stops and asks you.

To answer you I have to process a lot of data: about 5 times the usual amount. It will take me a little longer and the query will cost more. Shall I carry on?

Below there are two buttons:

The permission applies to that question. If you then ask something else that also brings a lot of data, it asks again.

It only asks about your data. When you ask it how to do something or where an option is, it never asks whether to continue: it reads the manual, the screens and the list of reports without asking your permission, and if it looks at some of your data along the way, it keeps what fits.

There is a limit that isn’t passed even if you say yes: anything excessive (tens of thousands of rows) is cut off, and the assistant tells you that the answer is incomplete and how to narrow it down. To get all the rows, use View as report.

Can I make the panel bigger?

Yes, in two ways:

It is worth it for reading a long analysis: at the small size the text runs in a narrow column.

Can I move the panel out of the way if it covers something?

Yes. Drag it by its header —the strip at the top, where it says ‘Assistant’— and take it wherever you like, like any window. The position is remembered for the next time you open it.

Clicking the header buttons still works the same: it only moves if you grab a gap between them. And if you shrink the browser window, the panel repositions itself so as not to end up off the screen.

At full screen it doesn’t move, because there it already fills the centre.

The assistant doesn’t close when you click outside it: it closes with its ✕. You can keep using the program with the panel open alongside, and the conversation carries on where you left it.

And if I miss the answer because I scrolled up to reread?

The thread follows the answer as it is written, however long it is, so you always have it in front of you without touching anything.

It stops following it as soon as you scroll up —with the wheel, by dragging the bar or with the keys—, because then you are reading something else and moving the text would be worse. It starts following it again on its own when you scroll back down to the end.

When that happens and the answer has been fully written, a notice appears above the box: Has replied · see the reply. Click it and it scrolls down to the end. If you scroll down by hand, the notice disappears on its own.

Can I copy the whole conversation?

Yes. In the panel header there is a copy button (⧉). It copies the whole conversation to the clipboard —your messages, the answers and also the action cards, with their details and how each one ended— ready to paste into an email, a support ticket or a document.

The button turns green with a ✔ for a moment to confirm it has been copied; you can’t see the clipboard, so it is the only sign that it worked.

Where does it get its answers from?

Answers about how the program works come from the Powergest documentation, not from the model’s memory. When the assistant finds documentation that backs up what it says, it cites the passage with a number in square brackets: [1], [2].

If it doesn’t find enough documentation, it says so and doesn’t make up the answer. This is deliberate: it prefers to admit it doesn’t know rather than give you steps that don’t exist.

As well as the product documentation, the assistant reads whatever your company has written in the knowledge guide. If that guide has an official answer to the question, that answer takes precedence over everything else, in your company’s own words.

When the answer is based on a help page, the citation carries a link to that page: clicking it opens it in Documentation, inside the program itself.

What is the ‘Correction’ that appears below an answer?

The program checks every menu path the assistant writes (‘Treasury → Direct debit batches’) against the real menu. If one isn’t like that, it corrects it below the answer:

Correction: in the menu it is Cost accounting → Dimensions, not ‘Accounting → Dimensions’.

If there is no screen with that path, it tells you so and invites you to ask it where it is. Trust the correction: it comes from the menu, not from the assistant’s memory.

Which financial year are the figures it gives me from?

From the dates you ask about, not from the financial year you have open on screen. Even if you are working in 2025, ‘how much have I invoiced this year?’ is answered with this year’s financial year, and ‘sales in September last year’ with last year’s. If you don’t give dates, it counts from today.

If the period you ask for spans two financial years (November to February, for example), it looks them up in parts or tells you that the figure is from only one of them. And if there is no financial year with those dates, it tells you there is no data, instead of telling you they are zero.

What language does it answer in?

In the one you have chosen in My settings → Preferences → Language. If you work in English, it answers in English even if you write the question in Spanish, and it gives you the names of screens and buttons as you see them.

Whatever is data isn’t translated: a customer’s name, an item’s description or what a colleague wrote come out as they are.

Links to the help also open in your language when the page is translated. The CRM and mailing ones are; the rest open in Spanish.

Can I ask it how much Powergest costs?

Yes. The assistant looks up the current price list and tells you which plans there are, how much they cost per month and per year —the prices it gives are excluding VAT—, how many artificial intelligence credits each one includes and what limits it has.

Credits are included in your plan’s fee. They can also be topped up by buying additional packs, but the sales team confirms their price: the assistant doesn’t give it.

On prices, fees and terms the assistant only says what is in the price list. If something isn’t there, it refers you to sales instead of venturing a figure. A made-up price isn’t a minor mistake: it is a commitment nobody has made.

How do I start a new conversation?

The assistant remembers what has been said in the current conversation. That is what lets you say ‘and that customer, how much do they owe me?’ without repeating the name. It also remembers which record each name it has already looked up corresponded to, so it doesn’t repeat the search or ask you for internal numbers; and a ‘yes, go ahead’ follows on from what was being done.

To start from scratch, click the New button in the panel header (it has a speech-bubble icon and the label New conversation). The thread is cleared and the next message starts a new conversation.

The previous conversation isn’t deleted: it stays saved. When you open the panel again, the last conversation is brought back with its messages, and you can carry on where you left off.

The ✕ button in the header only closes the panel. It doesn’t delete anything.

Can I dictate the question instead of typing it?

Yes. Next to the text box there is a microphone button with the label Dictate by voice.

  1. Click the microphone. It starts recording and a timer appears.
  2. Say your question.
  3. Click the microphone again, which now says Stop and transcribe.

The transcribed text goes into the box, it isn’t sent on its own. You can correct it before clicking Send. If you had already typed something, the dictated text is added after it. The recording isn’t stored anywhere.

Who transcribes what I dictate?

The microphone is in the assistant panel (floating button → AI assistant), next to the typing box. Who transcribes depends on the browser:

If neither is possible, the microphone is switched off and, when you hover over it, says why. Use Chrome or Edge, or type the question.

What you seeWhat has happenedWhat to do
The microphone is grey and can’t be clickedThe browser can’t transcribe and there is no AI machine at the siteOpen the ERP in Chrome or Edge
‘The browser won’t let the microphone be used…’You said no to the permission, or it is blockedClick the padlock in the address bar and allow the microphone
‘No microphone can be found on this computer.’There is no microphone connectedConnect one, or type the question
‘The browser’s dictation needs an internet connection…’Chrome and Edge transcribe on their serversCheck the connection
‘Nothing was heard…’There were a few seconds of silence and the browser stoppedClick again and speak close to the microphone

How do I say whether an answer was useful?

Under each assistant answer there are two buttons: 👍 (Useful) and 👎 (Not useful).

Clicking one records your rating. You can change it afterwards: the last one is kept.

Negative ratings are the main source of improvement for the assistant. Whoever administers the ERP reviews them and corrects whatever is needed in the documentation or in the knowledge guide. Marking 👎 without a comment already helps; a comment helps more.

Rating the last answer is also useful for something else: the assistant keeps the shape of that question. With a 👍 on an explanation taken from the manual, the next similar question finds those same pages sooner. With a 👎, it stops trusting what it had used to answer it. What it keeps is the shape of the question, without names or figures, and it stays in your company.

What if the assistant can't solve something about the program?

After a 👎, or when the assistant itself admits it doesn't know, Shall I pass it to support? appears under its answer. Press Pass it on and the conversation reaches Infoser as a request, with the screen you are on: no need to explain it again. It tells you the number, and the reply reaches you in Main → Help → Support, under My issues. If you don't want to, No, thanks.

If something goes wrong

What you seeWhat has happenedWhat to do
‘The assistant does not have an AI engine available yet…’Nobody has set up your installation’s AI engineYour installation’s technical service sets it up, in System → AI centre
‘The assistant is not turned on for this workspace…’Nobody has turned on the assistant in your installationAsk whoever administers the ERP
‘The assistant is not available right now: no AI node at the site is connected’The machine that serves the assistant is turned off or offlineTell support; it isn’t something you need to deal with
‘The shared assistant is not available right now. Try again in a few minutes’Momentary outage of the shared serviceTry again later
‘You have reached the daily query limit of the shared assistant for this workspace’Your company has used up its daily query quotaWait until tomorrow, or ask for a machine of your own
‘The company has used up its monthly AI budget…’The month’s budget has run outSpeak to your administrator
‘I don’t have enough information in the documentation…’There is no documentation to back up the answerRephrase it, or complete the knowledge guide
‘I could not complete the response due to a temporary AI service failure…’The AI engine was set up and failed when dealing with that particular questionAsk again. If it keeps happening, tell support: it isn’t down to you or your settings
‘The reply could not be obtained.’The connection was cut off halfwayAsk again

See also: Choose the thinking level · What the assistant knows and what it doesn’t · Ask it to do something · Attach an image or a PDF

More about Assistant