VISEON: vCat for Qlik Cloud Analytics
RISK: AI executes reporting and operations without knowledge and true business context.
VISEON for Qlik or vCat for short is the top-level navigation system for enterprise AI for your Qlik Cloud Analytics tenant. It maintains the canonical master map of your tenant—your live reporting, master items, and operating rules—ensuring AI agents know and understand your estate before they act, safely directing them to execute with zero guesswork and total auditability.
The gap data access doesn’t close
Data access tells an AI agent what exists. It doesn’t tell the agent what things mean.
Ask an AI agent a business question and it has to travel a path: read the intent behind the question, resolve what your business terms actually mean, identify the right metric, choose the correct implementation, then run the calculation. Tools that give agents data access — a Qlik MCP, a SQL agent — are good at the last two steps. They are not built to do the first three: understanding intent, resolving meaning, and picking the metric that actually answers the question. That’s the semantic gap, and it’s where most AI-on-analytics projects go wrong — not at the compute, but before it.
From your analytics estate to governed concepts
vCat is deployed at the edge, close to your tenant, and built as two products working together: an extension, which scans your Qlik Cloud Analytics estate — the master measures, dimensions and fields already built into your apps — and automatically drafts descriptions for them; and a portal, where your data steward reviews each draft. vCat proposes; your steward approves. Nothing is published to the estate as a governed description until a human has signed it off.
That matters because Qlik Answers already does the right thing with master item descriptions when they exist — it prioritises them over its own inference. The gap in most tenants isn’t the prioritisation logic, it’s the descriptions themselves: writing a clear, correct description for every master measure and dimension by hand doesn’t scale. vCat is what closes that gap — an automated, human-approved way to keep those descriptions complete, so the governed source Qlik Answers already trusts most is one your estate can actually keep up to date.
In our own testing, this made a measurable difference. Before governed descriptions were in place, Qlik Answers’ Thinking mode would occasionally fall back on the model’s training data rather than your data. After they were written back into the app, it stayed inside your data. That’s evidence, not a claim — and it’s the kind of before/after we can show you against your own tenant.
A business concept is not a formula
A concept like Net Revenue is a governed signpost, not a calculation. It carries a business meaning, a semantic identity that tells an AI agent how to reason with it, a governance state, and a pointer to one or more technical implementations. The concept says what is meant; your Qlik apps compute it. Kept separate, one governed meaning can point to several implementations across your estate — without an agent ever guessing which one you meant.
Context first, implementation second
At runtime, vCat and Qlik do two different jobs, in a fixed order. A user asks a question — “compare revenue against budget.” vCat MCP resolves intent, meaning, and the right metric, and returns a definition, a trust level and how fresh that definition is. Only then does Qlik MCP identify the implementation and run the computation, returning the actual number. The response an agent hands back carries all of it: the figure, the governed definition it’s based on, and the trust level behind it — so the answer comes with its own provenance, not just a number.
Today that ordering is enforced by how the skills are built; a dedicated orchestration layer sits on the roadmap. Either way, the principle holds: an AI agent should understand what you mean before it decides how to compute it.
Where this sits alongside Qlik Answers and the Qlik MCP Server
vCat isn’t a replacement for either — it’s the layer that makes both work smarter with what you already have.
Qlik Answers already prioritises your master item definitions over model inference, and cites the fields and expressions behind every answer. That’s sound design — but it only holds up where a description exists. Where master items are undescribed, Qlik Answers falls back to unstructured content and general LLM reasoning, which is where governed answers slip back toward guesswork. vCat’s job is to make sure that fallback is rarely needed, by keeping descriptions complete, approved and current across the estate.
The Qlik MCP Server is the execution layer: it gives an AI agent governed access to your apps, data products and datasets, runs the analytics engine, and returns a computed answer — steps four and five of the path above. vCat MCP sits ahead of it in the same request: resolving intent, meaning and the right metric before Qlik MCP is asked to compute anything. One worked example — “compare revenue against budget” — resolves through vCat first and executes through Qlik MCP second, with the response carrying both the number and the definition it’s based on.