BigQuery Graph turns agent memory into a query

What it does

Google Cloud on Tuesday made BigQuery Graph generally available, and the positioning is the tell: not faster analytics but “the knowledge foundation for the agentic era.” The premise is that the questions that matter in enterprise data are rarely about individual rows — they are about how things connect: how two accounts are linked, what path a payment took, what context grounds an AI agent’s answer. Historically that meant extracting data into standalone graph databases, with the silos and ETL overhead that follow. BigQuery Graph puts native graph capabilities directly in the warehouse: ISO-standard Graph Query Language sits alongside SQL, traversals run natively, and the graph inherits your existing row- and column-level security and can call BigQuery ML and AI functions in the same query — one engine doing petabyte-scale graph analytics and connected context for agents without the memory bottlenecks of a scale-up database. The GA release aims squarely at agents. A context graph in BigQuery Agent Analytics captures every action an agent takes and shapes it into a typed, queryable trace of the agent’s reasoning, stored in BigQuery Graph — so “why did the agent do this?” is a single traversal, and the outcomes you join back to those decisions become the data that improves the next one. Google has also packaged graph expertise itself as an agent skill — GQL pattern matching, blending graph and SQL, schema design — accessible in Antigravity, Visual Studio Code, Claude Code, and Codex through the Data Agent Kit extension, and Gemini Enterprise can connect to a graph through an MCP server. On the engine side, GQL runs 2x faster than preview against public benchmarks with undirected traversal up to 100x faster; a new CALL statement and extended subquery support let a graph query break into reusable parts, so the same functions an analyst writes become the building blocks an agent calls as a tool. The graph’s reach expanded too: with borderless lakehouse, one graph can span native BigQuery tables and open Iceberg tables in other clouds — Databricks Unity Catalog, AWS Glue, Snowflake — traversed in place, no copying and no ETL. For teams that don’t write GQL, conversational analytics chats with the graph in natural language, reading the relationships in your schema to translate questions into SQL or GQL and visualizing the traversal — the relationships themselves doing the grounding that cuts ambiguity and hallucination. Codelabs cover fraud detection, identity resolution, and supply-chain digital twins, plus one for the agent context graph itself.