OpenLLMetry is Traceloop’s open-source observability toolkit for generative AI applications, built on the OpenTelemetry standard. It instruments LLM calls, vector database queries and agent frameworks so every request your application makes to a model becomes a trace you can inspect. The project has about 7,500 stars on GitHub, is released under the Apache 2.0 license, and was actively developed in 2026.
When an AI feature misbehaves, “the model said something weird” is not a debugging strategy. OpenLLMetry extends OpenTelemetry — the observability standard your existing stack already uses — with semantic conventions for GenAI: prompts, completions, token usage, model parameters and costs are captured as structured spans and metrics. Out-of-the-box instrumentations cover providers such as OpenAI, Anthropic and Cohere, frameworks like LangChain, and vector stores including Pinecone and Chroma. Because the output is standard OpenTelemetry data, it flows into whatever backend you already run — Jaeger, Grafana, Datadog or Traceloop’s own platform — instead of locking you into yet another silo.
Freemium. OpenLLMetry itself is free and open source under the Apache 2.0 license. Traceloop offers a paid hosted platform with LLM-specific dashboards, evaluations and monitoring on top of the collected data.
Platform and application teams running LLM features in production who need real observability — latency, cost and behavior per request — inside the monitoring tools they already trust.