Skip to content

Monitoring and Observability ​

The Agent Starter Pack offers robust options to monitor your agent's behavior, performance, and resource usage. This allows you to debug issues, optimize costs, and understand how your agent is functioning.

There are two primary observability features available:

  1. Agent Telemetry Events (Cloud Trace): Provides OpenTelemetry-based tracing and spans for all agent operations, automatically exported to Google Cloud Trace. This is enabled by default and great for understanding execution flow and latency.

  2. BigQuery Agent Analytics Plugin: An opt-in plugin for ADK-based agents that logs detailed agent events, including LLM interactions, tool calls, and outcomes, directly to BigQuery. This enables in-depth analysis, LLM based evals and observability, use of BigQuery conversational analytics and custom dashboards.

Choosing an Approach ​

FeatureCloud Trace TelemetryBigQuery Agent Analytics Plugin
EnablingEnabled by defaultOpt-in via --bq-analytics flag
Primary Use CaseExecution flow, latency, debuggingDeep analysis, llm-as-a-judge, conversational analytics, dashboards
Data DestinationGoogle Cloud TraceGoogle BigQuery
SchemaOpenTelemetry SpansPredefined BigQuery Table Schema
Content LoggingSpan attributes (Metadata)Detailed JSON payloads, GCS offloading for large content
Agent CompatibilityAll TemplatesADK-based agents only
SetupNoneEnable flag during project creation

Recommendation:

  • Use Cloud Trace Telemetry for real-time debugging and performance monitoring of all agents.
  • Enable the BigQuery Agent Analytics Plugin when you need to perform detailed analysis on agent behavior, run LLM based evaluations, use BigQuery conversational analytics, track events over time, or build custom reporting dashboards for your ADK-based agents.

Released under the Apache 2.0 License.