Query BigLake data with natural language

This guide shows you how to use conversational analytics in BigQuery to query data in your BigLake tables with natural language prompts. By leveraging Google CloudBigLake, you can interact with your own data in BigLake as if it were a standard BigQuery table.

How conversational analytics works

Conversational analytics uses large language models (LLMs) to understand your natural language questions and map them to the schema of your BigLake tables. The process follows these steps:

  1. Schema discovery: The system retrieves metadata from the BigLake metastore to understand table structures, column names, and data types.
  2. SQL generation: The LLM generates an SQL query that is compatible with the BigQuery engine and the underlying data format.
  3. Execution: BigQuery executes the generated SQL query directly against the open-format data in BigLake.
  4. Response: The results are returned to the conversational interface, often accompanied by a summary or visualization.

For more information about conversational analytics, such as managing data agents, pricing, or best practices, see Conversational analytics overview.

Supported formats

Conversational analytics translates your natural language questions into SQL queries. It supports the open table formats supported by the BigLake metastore, such as Apache Iceberg tables.

Before you begin

Before you can query your data, register your external tables in the BigLake metastore. The BigLake metastore acts as the unified hub that connects BigQuery Studio to your external open-format data. Once connected, the tables become discoverable assets within BigQuery.

Query tables with conversational analytics

  1. In the Google Cloud console, go to the BigQuery Studio Agents Hub.

    Go to Agents Hub

  2. Create a Data Agent or start a direct conversation with an existing data agent.

  3. Select your BigLake tables.

    Because the BigLake metastore unifies all these different formats, the discovery experience is identical to finding standard BigQuery tables.

    1. Search: When you add your knowledge source, look up your table names in the table search and selection interface. You can use search keywords to filter results, including:

      • TABLE_NAME
      • catalog: CATALOG_NAME
      • project: PROJECT_ID
      • namespace: NAMESPACE_NAME

    2. Verify the source: Pay attention to the dataset portion of the fully qualified name. BigLake tables created by external sources and managed by the BigLake metastore will typically follow a format combining the catalog and namespace. For example: PROJECT_ID.biglake_catalog.finance_namespace.my_iceberg_table or PROJECT_ID.sap_catalog.sales.delta_table.

    3. Select: Add the selected BigLake table to your active conversational context.

  4. Ask questions in natural language. The system automatically translates your prompt into a federated SQL query.

Improve query accuracy

To help conversational analytics better understand your BigLake schemas and terminology, use Data Agent configuration options. These options include business glossaries, verified SQL queries, and system instructions.

What's next