Asking questions

How to get useful answers from Flux Intelligence, what a good question looks like, and how to read the response.

Flux Intelligence is an AI agent that answers questions about your estate. It plans its own investigation — choosing which governed tools to call and in what order — over the same data the dashboards use.

Where to find it #

Ask Flux is a panel available throughout the product. It already knows the page and filters you are looking at, so “why is this one high?” usually works without restating context.

The Intelligence Workspace is the full-screen surface for longer investigations.

Both need the role you already hold. Both read the same published snapshot the rest of the product does, so the assistant cannot tell you something the dashboard would contradict.

What makes a good question #

Flux Intelligence is strongest on questions that are analytical, bounded, and about your data:

Why did spend increase last Tuesday?
Which VMs are the best right-sizing candidates, and what evidence supports it?
What is my commitment utilization, and where is capacity going unused?
Which subscriptions have the worst tag coverage by unallocated spend?
What changed in the estate between Monday and Wednesday?

Less useful are questions that are unbounded (“tell me everything about my cloud”), or that ask for something Flux has no engine for — see How the assistant is bounded.

Narrow beats broad “Which VMs in production are oversized?” gets a sharper answer than “how can I save money?” The agent has a finite budget of tool calls; a scoped question spends it on depth rather than breadth.

Reading an answer #

Every response is structured rather than free prose:

Part What it is
Summary The short answer
Detail Tables, charts, and explanation
Facts What was actually retrieved
Interpretations What was inferred from it
Limitations Coverage, freshness, or uncertainty
Sources Which governed tools produced the answer
Follow-ups Useful next questions

The split between facts and interpretations is the part worth reading. It marks the boundary between what the data says and what the model concluded from it.

Partial coverage comes first If a source was throttled, stale, or incomplete, the answer opens by saying so and quantifying the gap. An answer that begins “Partial coverage:” is telling you something important about how far to trust the rest.

Answers include links back to the Flux page that proves the claim — the inventory view, the cost report, the opportunity. Nothing has to be taken on faith, and following the link is the fastest way to verify.

Charts and diagrams #

The assistant can return charts, but only from data a governed tool actually returned. It cannot draw a chart from numbers it invented, because it has no path to numbers that did not come from a tool.

Retention #

Prompts and validated replies are retained for a configurable window — typically 30 days — so administrators can review answer quality. Model reasoning is never retained.