The assistant
An optional Ask AI panel in the console. Ask a question in plain language and it looks the answer up in your own data sets, time series and events, then offers to open what it found in the console.
It is read only by design: the tools that change anything are never offered to it. It also acts as you, so it can only see the data sets you can see.
It is off until an administrator turns it on, and it needs a model, which costs money.
What it is good at
It is a way to ask a question without first knowing where the answer lives.
- Finding things. "Which data sets mention the compressor?" is faster to type than it is to browse for, especially in a large model.
- Counting and grouping. "How many alarms yesterday, by type?" It can group events by type, source, data set or day, then drill into whichever bucket matters.
- Following a thread. It can walk the knowledge graph outward from an asset to the time series that measure it, which is the question that usually starts an investigation.
- Handing off to the console. When an answer came from a lookup you could have done by hand, it offers a button that opens the same view, filtered the same way. The chat is where you ask, the console is still where you work.
What it will not do
It cannot change anything. That is not a policy in the prompt, it is enforced twice. Only read-only tools are offered to the model, so it cannot propose a change, and any request for a tool outside that list is refused before it runs.
It also inherits your permissions rather than holding its own. Every lookup is made as the signed-in user, so data set permissions apply exactly as they do everywhere else. Two people asking the same question can get different answers, and that is correct.
It is not a reporting tool, and it does not remember anything between conversations. Clearing the conversation discards it.
Effort: how hard it thinks
Below the message box is an Effort control with five settings, from low to maximum. It applies to the next message you send, and stays where you left it.
Higher effort means the model thinks longer before answering, and is willing to make more lookups on the way. That buys better answers to genuinely hard questions, and costs more time and more money on easy ones.
| Setting | Suits |
|---|---|
| Low, Medium | Lookups and counts, where the work is finding the data rather than reasoning about it |
| High | The default, and the right answer for most questions |
| Very high, Maximum | An investigation worth waiting for |
A rough rule: if you would have found the answer yourself in under a minute, low or medium will do. Maximum on a simple question mostly buys a longer wait.
Effort is the main thing that moves the bill, so it is worth knowing that an administrator can cap it. If a ceiling has been configured, no setting on this control can spend past it, and a long answer may be cut short instead. That is deliberate: a truncated answer is visible, an unexpected invoice is not.
If you cannot see it
It is off until an administrator turns it on, and that is three separate decisions: one for the deployment, one for your organisation, and one for you personally. If Ask AI is not in the top bar, ask your administrator which of the three is missing. Which model it runs on, hosted or inside your own network, is also their choice. Enabling the assistant →
- Enabling the assistant: the switches, the model and the settings, for administrators
- Data set permissions: what the assistant can and cannot see, since it inherits exactly your access
- Relationship analysis: the statistical companion to asking questions, for when the question is which signals move together
- Events: the data behind most of the counting questions people ask it