Using Topics in Amazon Quick chat
| Applies to: Enterprise Edition |
You can use Topics in Amazon Quick chat to ask natural language questions that span multiple datasets. The LLM-powered chat agent uses your defined relationships, dataset enrichment metadata, and custom instructions to generate cross-dataset SQL queries and return unified answers.
How cross-dataset chat queries work
When you ask a natural language question against a Topic, the chat agent performs the following steps:
-
Intent parsing. The agent identifies which columns map to your terms by matching against column names, descriptions, and synonyms from the enrichment metadata. It determines which datasets contain the relevant measures and dimensions.
-
Relationship traversal. Using the defined join keys and custom instructions, the agent determines the join path between identified datasets. It can traverse relationships to connect fact tables to the necessary dimension tables.
-
SQL generation. The agent constructs a SQL query with appropriate JOIN clauses, aggregation, and GROUP BY for the requested dimensions.
-
Result presentation. The answer is returned as a visualization or table, with the generated SQL available for inspection in the Explanation panel.
Chat capabilities with Topics
The LLM-powered chat agent supports richer SQL generation than the defined relationship path used in analysis sheets. When you configure a Topic for chat with custom instructions, the agent can generate:
-
Inner, left, right, and full outer joins
-
Union queries across tables with the same schema
-
Subqueries for negation patterns (for example, "customers who have never ordered")
-
Cross-grain comparisons (for example, daily actuals versus monthly targets)
-
Self-joins for recursive hierarchies (for example, employee-manager relationships)
The richer your enrichment metadata and custom instructions, the more accurately the agent interprets ambiguous questions.
Starting a chat with a Topic
To chat with a Topic:
-
Navigate to the Topic and choose the chat icon. Your Topic is automatically available as the context.
-
Or, from Amazon Quick chat, use the data filter to select Specific data and apps and choose your Topic.
Ask natural language questions that span multiple datasets. For example:
-
"Show total sales by customer segment and store region"
-
"What is the return rate by product category?"
-
"Which stores are below 80% of their monthly target?"
Use the Explanation panel to view the generated SQL and verify that the correct datasets and joins were used.
Differences from legacy Topics in chat
The chat experience differs significantly between new Topics and legacy Topics:
| Aspect | New Topics | Legacy Topics |
|---|---|---|
AI model |
LLM-powered chat agent |
Legacy ML-based fuzzy search model |
Cross-dataset queries |
Yes — traverses relationships and generates cross-dataset SQL with runtime joins |
No — selects one dataset and queries only that dataset |
Join types |
Inner, left, right, full outer, union, subquery |
Not applicable (single dataset) |
SQL visibility |
Generated SQL shown in Explanation panel |
Not available |
Note
You can chat with both new Topics and legacy Topics. However, legacy Topics use the legacy ML-based model and do not support cross-dataset queries. For information about legacy Topics, see Working with legacy Topics.
