Serve live, governed data in AI-built apps with Amazon Quick

Amazon Quick has introduced Live Data in Apps, a feature that lets published apps built by its AI agent query governed Amazon Quick Sight datasets live each time a user opens them, rather than showing numbers frozen at build time. The announcement was made in an AWS Machine Learning Blog post.

Previously, Quick Apps could bring live data from connectors and content sources such as action connectors (including Jira, Slack and Google Drive), Spaces documents, web search and AI inference, all running at view time. Governed Quick Sight datasets, including SPICE and Direct Query tables, could not be queried live, so any dataset numbers shown in an app were baked in when the agent built it.

Under the new feature, a builder describes the app in natural language; the agent finds relevant curated datasets, writes the SQL during the build, and asks the builder to approve each discovered dataset by name. After publishing, the app re-runs that SQL every time someone opens it, and the query executes as the person viewing it, so each user sees only the data they are allowed to see.

Prerequisites include access to datasets created with or without row-level security (RLS) and column-level security (CLS); both SPICE and Direct Query dataset modes are supported. Each viewer must consent per dataset on first use, and viewers must be authenticated Quick users — anonymous or public access is not supported. The minimum role for a builder or user is Reader Pro (Professional).

Existing RLS and CLS rules apply automatically, consent is enforced server-side on every query, and the Quick Sight query engine enforces authorization against each viewer's identity. Query and result-size guardrails surface a "narrow the query" message rather than truncated data. SPICE datasets or Direct Query datasets from the same data source can be used in a single app, but Direct Query datasets from different sources cannot be combined. Renaming or removing columns requires rebuilding queries in the app, and there is an initial row limit during build and row retrieval. The post also describes a sales-renewals use case and a view-time consent prompt.