Langtrace is an open-source observability tool for collecting and analyzing traces and metrics to improve the performance of large language model (LLM) applications. It supports OpenTelemetry standard tracing, can be self-hosted to avoid vendor lock-in, and offers end-to-end observability. Langtrace enables users to gain comprehensive insights into the entire machine learning process, including RAG or fine-tuned models. Additionally, Langtrace supports the establishment of feedback loops by creating gold datasets from traced LLM interactions to continuously test and enhance AI applications.