Hello Google Cloud team,
I’m currently working with a Gemini Enterprise (AgentSpace) setup where a large number of documents stored in Cloud Storage have been ingested into a Data Store. The Data Store is connected to AgentSpace, and I’m using the default agent to perform semantic search across these documents.
To further improve retrieval accuracy, I’d like to understand how metadata can be leveraged within this architecture, especially when combining BigQuery metadata tables with Cloud Storage content.
I’m referring to the following documentation:
According to the documentation, I understand how to register a Data Store using BigQuery metadata and Cloud Storage documents.
However, I’d like to clarify how the default AgentSpace agent actually uses that metadata during chat interactions.
Specifically:
-
When a Data Store includes metadata fields (e.g., category, language, created_date, etc.), does the default Gemini Enterprise agent automatically consider those fields when performing retrievals?
-
Or does the agent simply perform a full-text semantic search across indexed content, ignoring metadata unless a developer explicitly defines a filter or rank_boost rule?
-
Are there default behaviors (for example, boosting by recency or language matching) that apply automatically to metadata fields if present in the Data Store schema?
Understanding this is critical for designing prompt strategies, whether we need to include explicit metadata filtering logic inside the agent prompt or rely on the built-in retrieval layer.
Thank you