Amazon Bedrock Agents now integrates with Arize AI to provide comprehensive observability for AI agent applications. The integration offers tracing capabilities that capture detailed execution paths, from user queries through knowledge retrieval and action execution. Developers can monitor key metrics like response latency, token usage, and function calling accuracy. The solution includes both enterprise (Arize AX) and open-source (Arize Phoenix) versions, with automatic instrumentation through OpenInference libraries. The post demonstrates setting up tracing, viewing captured data in Phoenix dashboards, and evaluating agent performance using built-in LLM evaluation templates for function calling accuracy.
Table of contents
Solution overviewPrerequisitesInstall required dependenciesConfigure the Phoenix environmentSet up an Amazon Bedrock session and agentRun your agent with tracing enabledView captured traces in PhoenixEvaluate the agent in PhoenixConclusionSort: