Operator chat

Operator chat is a box on a Spectrum or Lens screen where an operator asks a question in plain language: why is Tank 1 alarming?, what changed on the pump house in the last hour? The answer is short and in plain text. It is decision support: the operator still decides what to do.

It needs more than the other surfaces:

• the AI Use capability for whoever is signed in, which Operator and Viewer roles do not have by default;
• the Operator chat switch on, and a license with both aiCore and aiOperator;
• someone to put the Iris Chat component on a screen in Studio.

What it does and does not do

It can read what the signed-in operator's role can read, through the same tools as every other surface. It is told which screen the operator is looking at and which tags are visible on it, as background.

It can offer exactly one action: propose a tag write, and only while AI may propose tag writes is on and the gateway's tag access is restricted. The proposal appears in the chat as a card showing the tag, the value and the reason, with the buttons the operator is allowed to use: only someone who holds both AI approve and the right to write that tag sees Approve. Everyone else sees that the change is waiting for somebody who can. See Approvals.

It cannot propose a change to a screen, script or any project file, and it cannot reach the designer's tools, whoever is asking.

An answer is plain text: the model is told not to use markup or links, and the screen shows whatever it sends as text, not as a link or a picture.

Kiosks with no sign-in

A kiosk that is open without a sign-in counts as one person, anonymous, as far as token and rate limits go: every such kiosk shares that person's limits, and each kiosk window has its own session allowance on top. Whether that role holds AI Use at all is your decision under Security › Roles.

Limits

Per session (one browser tab or Lens window): 10 requests a minute and 100,000 tokens a day, both changeable under Security › Iris AI › Operator chat limits. The person's own limits apply on top. A refused question says which limit it hit and when it clears.