Introducing Tracy: The AI Observability Library for Kotlin | The Kotlin Blog

Tracy is an open-source library designed to enhance observability in AI applications built with Kotlin, facilitating debugging, performance measurement, and tracking of language model (LLM) usage. By integrating seamlessly with popular Kotlin/LLM stacks and relying on OpenTelemetry, Tracy offers flexibility in exporting trace data to various backends like Jaeger or Grafana, and integrates with LLM engineering platforms such as Langfuse. Unlike full AI frameworks that limit tracing to framework APIs, Tracy permits detailed monitoring of LLM usage through API or HTTP client instrumentation and annotating Kotlin functions, allowing developers to trace application logic and LLM call metadata efficiently. It simplifies observability by providing high-level APIs for scoped spans, LLM client instrumentation, and tool call tracing, requiring minimal code changes. As an open-source project, Tracy invites community contributions to extend its capabilities and aims to support the growing Kotlin AI ecosystem by ensuring robust observability of AI applications.