How to trace Strands agents running on Amazon Bedrock AgentCore with Braintrust

Strands agents deployed on Amazon Bedrock AgentCore can export OpenTelemetry traces to Braintrust while sending model inference requests to Amazon Bedrock, capturing agent invocations, event cycles, model calls, tool executions, messages, and token usage. Reliable tracing requires initializing StrandsTelemetry once per runtime process before the agent handles requests, configuring Braintrust’s OTLP endpoint, authorization key, and stable project ID through environment variables, and preventing AgentCore’s default ADOT auto-instrumentation from registering a competing tracer provider. Deployments should disable trace auto-instrumentation, clear conflicting ADOT settings, securely manage and rotate Braintrust API keys, and ensure the runtime role can invoke the selected Bedrock model. Trace delivery can be verified by submitting distinctive tool-using requests and inspecting Braintrust’s trace hierarchy, inputs, outputs, model and tool spans, durations, and token metrics, while avoiding double-counting aggregate and individual usage. Common failures include incorrect endpoints or headers, provider-registration conflicts, initialization after the first request, oversized OTLP payloads, missing root spans across distributed services, and unflushed spans during shutdown. Braintrust can also support ongoing debugging, production issue classification, regression evaluation datasets, and asynchronous quality scoring, while a native Strands integration offers an alternative to the OpenTelemetry export path for new implementations.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.