Google Cloud has announced general availability of LookML semantic language indexing in Knowledge Catalog for Looker Core on Google Cloud. The release addresses a gap enterprises face when moving AI agents from prototypes to production: access to data is not the same as understanding it. Agents need both broad organizational context — table schemas, lineage, governance policies, business definitions — and deterministic business semantics such as exact SQL formulas, join paths, and filtering rules for metrics like net revenue or churn.

Knowledge Catalog provides the unified context graph across structured tables and unstructured documents, while LookML encodes production-tested semantic logic including multi-grain join protections, conditional filters, and KPI formulas. Without explicit semantic rules, text-to-SQL agents fall into metric definition drift, inventing conflicting formulas and ignoring rules like recognizing revenue only when order_status = ‘Complete’, and naive join fan-outs, multiplying rows across one-to-many relationships and inflating financial totals.

What’s new: LookML‘s deterministic semantics become a first-class layer in the Knowledge Catalog context graph. Previously, catalog integrations only captured shallow, file-level pointers and basic dashboard inventories. Now an agent that finds a metric also finds its exact definition, join path, and lineage back to source tables. Synchronization is always on by default, and live sync reflects production LookML updates in near-real-time. Because the metadata is indexed directly into Knowledge Catalog, it is immediately available to Gemini Enterprise without custom agent code, and to custom agents via the LookupContext API.

The integration captures multiple entity levels: LookML models with include hierarchies, data validation tests, and database connection routing; LookML Explores with full join logic, cardinality, mandatory filters, and starter queries; LookML Views and fields with compiled SQL expressions, primary keys, dimension timeframes, synonyms, and derived table logic; and dashboards, Looks, and tiles with filters, chart types, and lineage from BigQuery tables down to individual dashboard tiles.

The article compares three agent architectures for a complex cross-domain prompt. Pattern 1, a baseline database agent, guesses raw SQL with no semantic or catalog context, lacks visibility into governance, and burns tokens in multi-turn debugging. Pattern 2, a context-grounded agent using Knowledge Catalog without indexed semantics, finds assets and policy documents but still suffers from calculation drift and join fan-outs. Pattern 3, a context plus semantic-grounded agent using Knowledge Catalog with indexed LookML, retrieves exact join paths and field formulas, applies the May–July fiscal calendar and status = ‘Complete’, prevents join fan-outs, and returns an interactive Open in Looker Explore link alongside the answer.

The two experiences are positioned for complementary jobs. Looker Conversational Analytics remains the dedicated BI assistant for users working inside BI workflows and asking questions scoped to curated Explores. Knowledge Catalog plus Looker powers enterprise-wide agents in Gemini Enterprise across federated Data Cloud sources like BigQuery, spanning structured tables, BI metrics, and unstructured documents. Enabling Knowledge Catalog gives built-in skills to discover semantics across hundreds of LookML models and pass join definitions and starter queries to the BigQuery connector.

Getting started requires confirming first-party metadata indexing for Looker Core on Google Cloud, enabled by default for new instances with no extra charge for storing Looker technical metadata. Users must enable the Cloud Dataplex API and Data Lineage API and grant roles/dataplex.viewer, roles/looker.schemaViewer, and roles/datalineage.viewer. Then enable Knowledge Catalog in Gemini Enterprise, either automatically in federated query mode or via Configurations > Assistant. After that, the agent connects using end-user credentials, so users only discover and query assets they have IAM permission to view.