Langfuse

The platform provides end‑to‑end observability for large‑language‑model applications, capturing hierarchical traces of each model call, tool invocation, and retrieval step. Users can filter traces by user, session, cost, latency, or custom metadata, and view metrics through dashboards and alerts that monitor performance and expense.

It also bundles prompt management, evaluation, and experimentation tools. Prompts are stored separately from code, allowing one‑click deployment, rollbacks, and side‑by‑side testing on real production inputs. Evaluations can be run with LLM‑as‑a‑judge, heuristic functions, or human review, and experiments can compare results across models or configurations. Human‑in‑the‑loop annotation workflows enable the creation of golden datasets directly from traces.

The software is open source under the MIT license, can be self‑hosted in minutes, and integrates with any language or framework via OpenTelemetry and over 80 integrations. It is positioned for teams building chatbots, AI agents, or other LLM‑driven products who need collaborative debugging, monitoring, and iterative improvement.