This tutorial demonstrates how to implement a fully local, cloud-free observability pipeline for AI agents using OpenTelemetry GenAI semantic conventions, Docker, and the Jaeger tracing platform.

Key Points

• Deploy a local OpenTelemetry Collector and Jaeger instance using Docker Compose with pinned image versions.
• Instrument Node.js AI agents with OpenTelemetry SDKs to capture model identifiers, token usage, and operation types.
• Use startActiveSpan to create parent-child hierarchies for multi-step reasoning chains and complex agent flows.
• Visualize trace data, latency, and token costs by building a custom React dashboard connected to the Jaeger API.
• Ensure data privacy by keeping all telemetry, prompts, and sensitive business logic within your local infrastructure.

Why it Matters

Implementing vendor-neutral observability allows teams to monitor AI performance and costs without exposing sensitive data to third-party cloud providers. This local-first approach ensures compliance while providing the granular visibility needed to debug complex, non-deterministic AI agent behaviors.