Langfuse combines tracing, prompt management and evaluations

Langfuse is an open-source, self-hostable AI engineering platform that helps teams collaboratively debug, analyze and improve agent applications. Its integrated tools cover observability, prompt and skill management, and evaluation, connecting development experiments with production monitoring.

Tracing captures LLM calls alongside retrieval, embeddings, API requests and other application logic. Teams can inspect conversations as sessions, visualize agent workflows as graphs, and monitor costs, latency and quality through dashboards and alerts. Data can be captured through native Python and JavaScript SDKs, framework integrations, OpenTelemetry or gateways such as LiteLLM. The OpenTelemetry foundation is intended to improve compatibility and reduce vendor lock-in.

Prompt management supports collaborative editing, version control and deployment through environment labels without code changes. Teams can test prompts interactively, compare performance across versions and link prompts to traces. Skill management, currently in beta, extends versioning and labels to agent instructions and supporting files.

Evaluation tools support production scoring and development testing through LLM-as-a-judge, code evaluators, user feedback, manual annotation and custom pipelines. Datasets and experiments let teams compare prompts and models, including through the interface without custom code. Public APIs and data exports support integration with broader engineering workflows.