This sample demonstrates how to build an agent that interacts directly with Google BigQuery, while also leveraging the BigQueryAgentAnalyticsPlugin for observability and logging.

• BigQuery Tools: The agent is equipped with native tools (e.g., execute_sql, list_dataset_ids, get_table_info, ask_data_insights) that allow it to explore available metadata, gain data insights, and execute SQL queries against BigQuery datasets.
• Analytics Plugin: Every interaction, tool call, and state transition is centrally logged to a BigQuery dataset (agent_analytics) for analysis, debugging, and auditing using the built-in ADK plugin.

1. A Google Cloud Project with the BigQuery API enabled.
2. Application Default Credentials configured (e.g., via gcloud auth application-default login).
3. Python >3.10.

1. Environment Variables: Create a .env file in the root of the application based on the .env.example:

   cp .env.example .env

   Then, update .env with your actual Google Cloud Project ID and desired location.

   Optional: You can also configure the BQ_ANALYTICS_DATASET_ID variable if you wish to write logs to a dataset other than the default (adk_agent_analytics).

2. Provision the Analytics Dataset: The ADK analytics plugin will automatically provision the underlying logging tables, but the dataset itself must be created first. Run the following command (substituting your dataset name and location if you changed them):

   bq mk --location=us-central1 --dataset "your-project-id:adk_agent_analytics"

3. Dependencies: Ensure uv is installed, then run the application using the starter pack CLI or direct ADK commands.

To test this agent interactively in a terminal UI:

adk run

Or, to launch the Web UI:

adk web

Ask the agent a question like:

"List the datasets available in my project." "Write and run a query to get the top 10 most popular programming languages on GitHub using bigquery-public-data.github_repos.languages."

As you interact with the agent, notice that a dataset named agent_analytics is automatically created (if it didn't exist) in your project, populated with detailed agent execution logs.