Quick answer
The Amazon Quick agentic catalog preview creates two different assets:
- Catalog-Generated Datasets: DirectQuery representations of selected catalog tables, with inherited table and column descriptions.
- Topics: multi-dataset semantic boundaries that use reviewed relationships to support Q&A across the selected Datasets.
The upstream catalog remains authoritative. The Quick Agent helps create representations; it does not certify that the table selection, definitions, or joins are correct.
Dataset behavior
AWS documents four current properties:
| Property | Meaning |
|---|---|
| Semantics inherited | The Dataset displays a badge showing definitions came from the upstream catalog |
| DirectQuery | Data is queried at the source by default rather than copied into SPICE |
| Sync semantics | An author can manually refresh table and column definitions |
| Customizable | Editing, transformations, or SPICE conversion creates a standard Dataset and ends semantic sync |
In the preview, inherited metadata includes table and column descriptions plus primary and foreign key definitions. Metric views and semantic views are described as future support, not current inherited behavior.
Read the semantic sync governance guide before customizing a generated Dataset.
Topic behavior
After Datasets are created, the Agent can inherit catalog relationships and recommend additional inferred ones. It can then assemble a Topic spanning the approved assets.
Topics are where join risk becomes visible. Review:
- fact and dimension grain;
- primary and foreign key uniqueness;
- relationship cardinality;
- null and orphan key behavior;
- multiple join paths and fan-out;
- time dimensions and slowly changing dimensions;
- default aggregations and metric definitions.
Validate with reconciled totals. A visually plausible star schema can still double-count revenue or omit unmatched records.
Build a focused context boundary
Create only the Datasets required for the intended business questions. A Topic that includes every discoverable table can introduce ambiguous joins, conflicting definitions, and irrelevant search candidates.
Document:
- intended users and decisions;
- approved questions and metrics;
- source owners and refresh expectations;
- excluded sensitive or irrelevant assets;
- relationship review evidence;
- sharing and export permissions;
- rollback and incident ownership.
The setup guide covers the discovery and creation sequence.
Test Q&A and dashboards separately
A Topic can support natural-language Q&A while its Datasets also support deterministic dashboards. Test both:
- fixed dashboard calculations and filters;
- paraphrased and ambiguous questions;
- multi-table questions across each join path;
- restricted users and denied data;
- stale or conflicting descriptions;
- source changes before and after semantic sync;
- exports and downstream reuse.
An answer that sounds grounded is not proof of a correct query. Inspect generated queries, source results, joins, and reconciled reference answers.
Sharing checklist
- Preview use is accepted for the environment.
- Every Dataset is intentionally included.
- Every inherited and inferred relationship has an owner.
- Known totals reconcile at each supported grain.
- Identity propagation or Quick RLS and CLS are tested.
- Semantic sync ownership and cadence are defined.
- Dataset customization triggers a fresh review.
- End users can report incorrect answers or definitions.
Frequently asked questions
What is a Catalog-Generated Dataset in Amazon Quick?
It is a DirectQuery representation created from a selected upstream catalog asset. It displays a Semantics inherited badge and keeps inherited definitions read-only while sync remains available.
What is a generated Topic?
It is a multi-dataset Topic built from selected Datasets and reviewed relationships, including inherited primary and foreign keys or additional relationships inferred by the Agent.
Are generated relationships guaranteed correct?
No. AWS requires authors to review inferred relationships. Reconcile keys, cardinality, join paths, grains, and known totals before sharing a Topic.
Official sources
- Amazon Quick: Behavior of created datasets
- Amazon Quick: The agentic catalog experience
- AWS launch post: Agentic Catalog Experience
Source check: August 4, 2026. Verify preview behavior, inheritance scope, DirectQuery, relationships, and sharing controls before use.