JetBrains Unveils AI Tracing Library for Kotlin and Java

Why It Matters

By providing standardized observability for generative‑AI workloads, Tracy helps teams monitor costs, debug model behavior, and integrate AI metrics into existing telemetry pipelines, accelerating reliable AI product development.

Key Takeaways

• Tracy provides OpenTelemetry‑compatible AI observability for Kotlin/Java
• Captures LLM token usage, cost, and execution time
• Supports OpenAI, Anthropic, Gemini SDKs out‑of‑the‑box
• Exports traces to Langfuse and Weave back ends
• Licensed Apache 2.0; works with Kotlin 2.0+ and Java 17+

Pulse Analysis

The rapid adoption of generative AI has outpaced the tooling needed to monitor its hidden costs and performance characteristics. Traditional application tracing focuses on request latency and error rates, but LLM‑driven features introduce variables such as token consumption, model latency, and API pricing. JetBrains’ Tracy arrives at this inflection point, offering a purpose‑built observability layer that aligns AI‑specific metrics with the broader OpenTelemetry ecosystem, allowing engineers to treat AI calls like any other service call in their monitoring stack.

Technically, Tracy leverages the OpenTelemetry Generative AI Semantic Conventions, ensuring that span attributes and event names are consistent across vendors. The library ships with ready‑made adapters for popular LLM providers—OpenAI, Anthropic, and Gemini—plus integration points for common Kotlin HTTP clients like OkHttp and Ktor. Developers can manually create spans, record inputs and outputs, and automatically capture token counts, execution duration, and monetary cost. Export destinations such as Langfuse and Weave mean that organizations can funnel AI traces into existing observability dashboards without building custom pipelines.

From a market perspective, Tracy lowers the barrier for enterprises to adopt responsible AI practices. By making cost visibility and debugging transparent, product teams can iterate faster while avoiding unexpected expense spikes. Its Apache 2.0 license encourages community contributions, positioning Tracy as a potential de‑facto standard for AI observability in JVM ecosystems. As more companies embed LLMs into core services, tools like Tracy will become essential for maintaining operational control and regulatory compliance.