Gain Full Visibility into AI Agents: BigQuery Agent Analytics Observability via Looker

Looker Tips & Resources Technical Tips & Tricks
1

As you build agents using the Google Agent Development Kit (ADK) or LangChain/LangGraph, the key to optimizing performance and managing token utilization is in deep observability. Understanding how your agents interact, identifying where they might be failing, and gaining clear insights into their operations are essential actions for driving improvements and ensuring transparency for billing and usage.

The BigQuery Agent Analytics Block, now available in the Looker Marketplace, provides an analytics solution designed to give you immediate visibility into your AI agent’s behavior and performance.

image
image1982Ă—1168 185 KB

Powered by Your Data in BigQuery

The Looker Block seamlessly connects to the telemetry data you collect using the open-source BigQuery Agent Analytics solution. If you’re using frameworks like the ADK or LangGraph, BigQuery Agent Analytics helps you stream rich event data (user requests, agent responses, tool calls, errors, etc.) directly into BigQuery. The Looker Block then sits on top of this data, transforming it into actionable insights.

What can you do with the Agent Analytics Looker Block?

This Block offers pre-built dashboards and Looker Explores to help you analyze:

  • Token Consumption: Keep tabs on key metrics like token consumption for the entire project and see a detailed breakdown of how LLM tokens are utilized across different agents, users, and timeframes. Tracks the volume of data processed to help manage costs and identify the primary drivers of resource usage.

  • Agents & Sessions: Quickly identify a detailed view of interaction volume, user engagement, and agent activity distribution. Track sessions, traces, and user growth to measure system adoption and operational load.

  • Tool Usage: Focus on specific functions and tools the agents invoke to complete their tasks.

  • LLM Interactions: Understand trends of the volume, costs, and patterns of your agent’s calls to Large Language Models.

  • User Analytics: Track user and client interactions with your agents.

  • Deep Dive into Interactions: Use Looker’s interactive features to drill into specific agent sessions or behaviors for root-cause analysis. The block features advanced visual drilling, allowing you to click any metric to launch context-aware pop-up visualizations.

Frictionless Setup

This block leverages Looker’s Native Derived Table architecture to parse JSON payloads on the fly directly within BigQuery. This means there is no need to build or maintain additional data pipelines.

During installation from the Marketplace, simply provide your BigQuery Project ID, Dataset Name, and the Base Table Name (default is agent_events).

Get Started in Minutes:

  1. Be sure you have BigQuery Agent Analytics set up to send your agent’s telemetry to a BigQuery table. (See the docs and codelab if needed).

  2. Install the “BigQuery Agent Analytics” block from the Looker Marketplace.

  3. Connect the block to your BigQuery instance, specifying the project, dataset, and table.

  4. Immediately start using the pre-built dashboards and Explores to analyze your agents!

This Looker Block, combined with the BigQuery Agent Analytics data collector, provides a powerful end-to-end solution for AI agent observability on Google Cloud.

Install the block, explore your agent data, and let us know your feedback in the comments below or by filing an issue on the GitHub repository.

The Agent Analytics Looker Block is open source! Check out the code, report issues, or contribute on GitHub.

2 Likes