Introducing new self-learning context enrichment in Knowledge Catalog

Google Cloud is introducing self-learning context enrichment in Knowledge Catalog, a preview feature that analyzes agent and user interactions to identify gaps in business context and keep definitions current. The system automatically extracts verified definitions from playbooks, code repositories, and business documents, creating reusable context sets. This replaces manual updates by context engineers and data stewards, who previously had to maintain definitions across thousands of assets. The enrichment capability connects to structured and unstructured data sources like Cloud Storage, Google Workspace, GitHub, Confluence, and SharePoint to assemble and maintain business knowledge.

Knowledge Studio, a reimagined interface in Knowledge Catalog, serves as a command center for context engineers and data stewards to explore data estates conversationally and navigate a visual context graph. It allows users to review proposed changes to the context graph before publishing them to agents and users. The system proposes business definitions, table descriptions, join relationships, golden queries, and agent instructions with source citations and confidence scores. It detects changes at the source, retires obsolete definitions, and flags conflicts for review, learning from both user and agent interactions.

The self-learning enrichment capability improves agent performance by nearly 4X in accuracy, reduces latency by up to 33%, and lowers token usage by up to 75%, according to internal benchmarks. It addresses challenges like ambiguous column lookups, ungrounded metrics, and missing definitions by pinpointing where agents struggle and staging fixes. Automation does not remove control; users can enable or disable auto-publishing, set confidence thresholds for automatic approvals, and manage sensitive updates through change requests. Every change is logged with provenance and can be reverted, and Google does not use customer data to train its models.

Context sets in Knowledge Catalog provide curated, governed subsets of the context graph tailored to specific domains, teams, or agents, such as Sales or Finance. Each domain maintains its own definitions, ensuring consistency within teams while allowing agents to dynamically search and retrieve the right context for any task. Context sets include built-in data governance and fine-grained access control via IAM, with traceable and reversible updates. They are portable through the Open Knowledge Format (OKF), enabling import and export of context bundles. The reimagined Knowledge Studio experience integrates these capabilities into an interactive command center for context engineers and data stewards.