AI tools
Six optional tools, governed by three independent switches. This page is the operator's view — what to turn on, where, and what it costs. What each tool does for a user is on the AI page.
On this page
Nothing happens by default#
AI is off. It requires both a switch and a configured key before a single call is made, and every product and screen type works identically without it. If your AI policy is not settled, you can deploy everything else and come back to this page later.
The three switches#
- The engine decides whether AI exists at all, which models may be used, and the daily ceiling. It runs in your own BTP subaccount, so this switch is yours in the most literal sense.
- The tenant turns individual tools on or off for everyone, in the settings dialog, and can pin a model per tool.
- Each screen can switch off the tools its own users should not have.
Each layer can only narrow the one above it. A screen cannot enable a tool the tenant switched off, and a tenant cannot enable one the engine does not offer — so the operator's decision is the ceiling, not a default somebody can drift past.
Turning a tool on#
- Confirm with whoever runs your engine that AI is enabled there.
- Open Settings → AI tools in the app. Each tool is listed with its own switch.
- Turn on the ones you want. The six are: Ask this screen, Explain this error, What is this field?, Screen-building copilot, Generate a template, Screen doctor.
- If a screen should not offer one of them, switch it off in that screen's settings.
Choosing a model#
The engine holds an allowlist. A tenant may pick per tool from that list and no further — a value that is no longer allowed shows as such rather than silently costing more, and a tenant cannot escalate itself onto a more expensive model than the operator permitted.
If only one model is allowed, the choice is not shown at all, because there is nothing to decide.
Watching the spend#
Usage is counted per tool: calls, tokens, model and an estimated cost, in the same settings dialog. There is a daily ceiling set on the engine.
Answers served from the knowledge base or the response cache are free and do not count against the ceiling, because they did not call anything — which is why the error explainer is cheap to leave on.
Turning it off again#
Any of the three layers switches it off, and the effect is immediate. Turning it off at the engine disables it for the whole tenant regardless of what anyone configured below — which is the lever to reach for if a policy question comes up mid-week.
Exactly what each tool sends while it is on is listed on the AI page and again on the security page. Prompts and answers are never stored by us.