Skip to content

How do I choose how much effort the assistant makes?

The *assistant*’s Premium AI has a thinking level selector with three positions: Fast, Normal and Deep. It is the only thing you decide about the engine. You never choose the specific artificial intelligence model: the program takes care of that.

Where is the selector and what does each level mean?

The selector only appears when the conversation is in Premium: because you have chosen it in the Model control, or because the assistant is offering you a professional analysis. In Automatic —the everyday basic AI— there are no levels to choose and they aren’t shown; below the bar it says that basic AI is included in your subscription and that it will suggest Premium if a question needs it.

When it is shown, it is in the bottom bar of the panel, to the right of the microphone: there are three buttons side by side (Fast · Normal · Deep) and the chosen one is highlighted in white, without opening anything. Below the bar it says what the chosen level means: the normal levels are included in your subscription for normal use, and Deep warns that it uses considerably more credits.

LevelWhat for
FastThe usual one. Questions about using the program, data queries, specific things
NormalQuestions that call for some reasoning or comparing several things
DeepAnalyses, diagnoses and decisions that weigh up many factors

The level applies to the question you send next, not to the whole conversation. You can change it between one question and the next.

The default is Fast. The program remembers your last choice in that browser, so if you usually work in Normal you will find it like that next time.

What really changes when I raise the level?

Raising the level does not make the assistant know more things. The documentation it consults and the company data it can reach are exactly the same.

What changes is the quality ceiling the answer can reach. The level sets how high it is allowed to go when you ask for a professional analysis. In Fast that ceiling is the lowest; in Deep, the highest.

In practice: raising the level doesn’t make anything more expensive by itself. What costs credits is asking for the professional analysis, and you always ask for that by hand —unless you have chosen in your preferences not to be asked, and even then the ceiling stays the same.

The level is a ceiling, not an instruction. Within it, the assistant uses the capacity the question needs: a simple query in Deep doesn’t use the maximum just because the selector is set there. What was actually used is shown by the label under each answer.

Put the other way round, which is what usually worries people: leaving the selector on Deep doesn’t make day-to-day use more expensive. Easy questions are still served cheaply.

Does the level work the same in every installation?

Yes. The level you choose and the decision about whether a query deserves more capacity are the same everywhere, whether or not your company has a machine dedicated to artificial intelligence.

That machine, if there is one, changes where the queries on your data are run —inside your network, instead of on the server—. It doesn’t change how much capacity goes into thinking about the answer or when you are offered a professional analysis.

In practice: if the selector doesn’t seem to do anything, it isn’t because your installation is missing a machine. Most likely the question was answered well below the ceiling, which is exactly what is wanted.

Why can’t I choose the artificial intelligence model?

Because the model is a technical decision that changes over time, and getting it right isn’t the user’s job. The program makes it based on what is asked, the level chosen and what your company has subscribed to.

That is why you will never see a model or provider name in the panel. Not even on the professional analysis card, which is called that precisely so as not to be tied to a brand name.

Whoever administers the ERP can see and change the engine, on the technical models screen. It is a restricted screen and isn’t in the menu.

Why are some levels missing from the selector?

The selector only shows the levels your profile is allowed. If someone sees three and you see two, your ceilings aren’t the same.

If you choose a level your plan doesn’t include, or the Premium credits have been used up, you don’t get an error: a window opens that tells you how to get it (Upgrade the AI, Buy credits or Change plan). It is explained in Ask for a professional analysis.

If what is missing is a level because of your profile, speak to whoever administers the ERP: it isn’t a fault in the screen.

If something goes wrong

What you seeWhat has happenedWhat to do
I can’t see the selectorThe conversation is in Automatic (basic AI)Click Premium in the Model control
The selector is greyed outAn answer is in progress, an action is pending or you are recordingWait for it to finish, or deal with the action
Only Fast appearsYour profile or your plan doesn’t go any higherCheck with whoever administers the ERP
I switch to Deep and the answer looks the sameThe level only raises the ceiling; without a professional analysis the answer comes from the usual engineAsk for the ✨ Professional analysis if you want the quality answer
I ask for Deep and the label says FastThe level is a ceiling: within it, whatever the question needs is usedIf you think more was needed, rephrase it with what you want it to take into account
I ask the same thing again and now it does offer me more capacityAsking again is taken as a sign that the answer wasn’t good enoughThat is intended. Accept if you want the quality answer; offering doesn’t charge anything

See also: See what level it answers at · Ask for a professional analysis · Ask the assistant something

More about Assistant