:kotlin_emoji:*<https://kotl.in/tracy-ai Introduci...
# announcements
m
K Introducing Tracy: Add observability to your Kotlin AI apps 🔍 Tracy, a library that adds production-grade observability to AI-powered apps in minutes, is now open-source. When you're building with LLMs, Tracy helps you unwind the timing of and causal relations between AI components without cluttering your codebase with manual instrumentation. Highlights: • Drop-in client support_:_ Use Tracy with OkHttp, Ktor, OpenAI, Anthropic, and Gemini clients. •
@Trace
annotations_:_ Track tool executions and your app’s internal flow cleanly, thanks to decorator-like simplicity now brought to Kotlin. • OpenTelemetry under the hood: Export to standard backends or dedicated LLM platforms like Langfuse. • Privacy first_:_ Rest assured that Tracy captures only metadata by default; data sharing for sensitive prompts requires explicit programmatic opt-in. Tracy also complements frameworks like Koog by adding visibility to your app’s internal flow alongside their built-in tracing. We’d love your feedback to help us shape the next iterations! 📎 Learn more: https://kotl.in/tracy-ai
K 23
d
Hi! It was not clear to me reading the article, whether this is an overlay above OTel SDK and how it overlaps with Kotlin SDK (https://github.com/open-telemetry/opentelemetry-kotlin) that has been given by Embrace to OTel (i.e. the CNCF)
Also, slightly OT, but could you guys give some news on this? https://youtrack.jetbrains.com/issue/IJPL-207858
v
@Didier Villevalois Hi! Good question.
Tracy
is a more high-level library compared to the
OpenTelemetry Kotlin SDK
. OpenTelemetry provides the core telemetry APIs/SDK, while Tracy focuses on AI-specific instrumentation and developer experience. For example, with Tracy you get out-of-the-box tracing for common AI providers/clients (OpenAI, Anthropic, Gemini), so LLM calls are traced automatically without writing custom instrumentation. We also support annotation-based tracing: you can trace functions with a simple
@Trace
annotation, with context propagation handled automatically and no manual span management. And in addition to standard OpenTelemetry backends, Tracy also supports LLM-focused observability platforms out of the box (e.g. Langfuse).
👍 1
d
I see. Thanks a lot for your answer! 😛
@Viacheslav Suvorov just a followup question. It looks like Tracy is JVM-only and does use the Java OTel SDK instead of the new Kotlin Multiplatform SDK. This does limit a lot its usefulness. Is it planned to migrate to the new OTel Kotlin SDK and make Tracy multiplatform?
v
@Didier Villevalois Tracy is currently JVM-focused and uses the Java OpenTelemetry SDK. Moving toward the Kotlin SDK and supporting KMP is definitely something we’re considering for the future.
K 1
❤️ 1