Knowledge Catalog of Google reads the metric rules of six tools

Google Cloud released a change in its Knowledge Catalog. The catalog reads semantic models from six data tools. A semantic model is a set of business rules, such as how to calculate net revenue. The catalog puts them in one map, and agents use the same rules as analysts. Google tells that, at this time, each agent must make a guess at these rules.

How it works

The catalog reads rules from these tools:

• Looker
• dbt MetricFlow
• Snowflake Semantic Views
• Databricks Metric Views
• SAP Core Schema Notation
• BigQuery Graph

It does not move data. When tools give different rules for one metric, the catalog shows each rule with its source. Data stewards then select the correct rule.

The numbers

In the internal tests of Google, agents with the catalog are almost 4X more accurate. They have a maximum of 33% less latency (wait time) and use a maximum of 75% less tokens (units of text that AI reads). Google gave no test data.

What is not known

No external test shows these numbers. The catalog also gets information from meeting records and chat threads, and Google gave no results for this part. Knowledge Studio, the tool to see the map, is in preview.

Insight: Google wants to hold the map of other data tools

In October 2026, Neo4j released a tool to make a layer of business data for agents. Also in October, Google added the rules of one more tool to its catalog. It keeps different rules of six tools side by side. This reads as a move to become the primary layer that agents read for business rules. That layer becomes the primary source. Check which layer will hold your metric rules.