Amazon Web Services (AWS) announced on June 22 at the AWS New York Summit the launch of its new service, 'AWS Context.'
AWS Context is designed to help artificial intelligence (AI) agents secure reliable data context. The service automatically maps relationships between existing data within organizations to create knowledge graphs and provides agent-based search capabilities.
Through AWS Context, the company aims to enable AI agents to utilize trustworthy context. The performance of AI agents is heavily influenced by the range of available context, which is often scattered across data lakes, data warehouses, lakehouses, databases, and data streams, as well as undocumented organizational knowledge.
Recently, major global tech companies have focused on improving the understanding of internal data by enterprise AI agents and their accuracy in performing tasks. AWS Context is seen as a solution to prevent AI agents from generating incorrect responses, a phenomenon known as hallucination, and aims to enhance business productivity through reliable, customized memory (context).
With AWS Context, AI agents within organizations can safely access data relationships governed by business rules and domain knowledge. Data stewards and curators can review inferred relationships in the console and incorporate domain-specific knowledge, such as business definitions and usage rules, to manage the graph.
This service extends the knowledge graph technology underlying Amazon QuickSight to an organizational level. The knowledge graph in Amazon QuickSight systematically connects and manages datasets, dashboards, and metadata while learning user patterns to continuously improve user experience. AWS expands this to provide a shared context layer that agents and applications within organizations can utilize.
The company explained that the more agents use AWS Context, the more sophisticated it becomes. During the querying process of the knowledge graph, agents identify which data sources yield accurate results, which join paths are predominantly used, and which curation rules are applied. If one agent discovers the correct join path or resolves schema ambiguity, other agents can leverage this without additional manual intervention.
Governance is also applied based on identity. All queries made through AWS Context are subject to the requester's AWS Identity and Access Management (IAM) and AWS Lake Formation permissions. Consequently, agents can only access and explore data and relationships permitted by their identity, allowing security and compliance teams to verify which data agents accessed and with what permissions.
AWS Context publishes key metadata for structured and unstructured data in the Apache Iceberg format within Amazon S3 tables. Users can query this context using Iceberg-compatible engines such as Amazon Athena, Amazon Redshift, and Apache Spark, enabling them to build, audit, or migrate systems based on this context.
AWS also previewed business context and semantic search features for the AWS Glue Data Catalog, Glue Data Catalog skill assets, and officially launched Amazon S3 annotation capabilities.
An AWS representative stated, "We define context as a data lake for AI agents and are building a knowledge and intelligence foundation for AI agents interacting with data."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.