Hello Google Cloud Community / Support Team,
We are currently building a data governance and conversational analytics setup using BigQuery Data Agent integrated with Dataplex Knowledge Catalog.
We have encountered a critical limitation where the Agent only imports and recognizes a maximum of 10 Glossaries from Knowledge Catalog, regardless of how many column-level links (EntryLinks) are configured.
Environment & Setup
-
Knowledge Catalog (Dataplex): Created a glossary containing over 1,000 Business Terms with detailed
DescriptionandOverviewattributes. -
Column-Level EntryLinks: Linked these terms to specific columns (e.g.,
ORDER_QUANTITY) in our target BigQuery table/view (DM_T_BUSINESS_SUMMARY_TRANSACTION_lite_copy_260710_copy). -
Dataplex Auto-Discovery: Asset Discovery is enabled and successfully running (
Succeeded). -
Verified Behavior: For the 10 terms that do load into the Agent, the Agent successfully reads both
DescriptionandOverviewattributes (including complex business logic andCASE WHENstatements).
Issue Description
Despite having hundreds of terms linked at the column level, the BigQuery Data Agent UI and backend context retriever strictly limit imported glossaries to 10 items (Glossary (10) / Imported from Knowledge Catalog (10)).
When querying the Agent about a column whose term is linked in Knowledge Catalog but falls outside these 10 imported items (e.g., ALLOWANCE_QUANTITY):
-
The background Context Retriever fails to fetch the linked Catalog Term.
-
As a fallback, the Agent attempts to run a SQL query against
INFORMATION_SCHEMA.COLUMN_FIELD_PATHS, which fails with:Error: Attempting to access tables outside the conversation's context definition. -
The Agent then falls back to general LLM inference rather than providing our curated business definitions.
Questions
-
Is there a default limit parameter (e.g., Top 10 / Preview Limit) on the Context Retriever API when pulling linked EntryLinks into BigQuery Data Agent?
-
How can we lift or adjust this 10-item limit so that the Agent can dynamically reference all column-linked Glossary Terms across our dataset?





