← All organizations

AI observability and evaluation

Langfuse

Langfuse builds an open-source platform for developers to debug and improve applications powered by large language models. Its tracing tools capture model calls, tool invocations and retrieval steps, letting engineers investigate how a request produced its answer. Teams can turn production examples into test datasets, compare changes through experiments and evaluate responses with code, model-based judges or human review. Prompt management lets them deploy and roll back prompts separately from application code. Canva’s AI team uses Langfuse to trace and debug generative design features in production. Teams can use Langfuse Cloud or host the platform on their own infrastructure.

Langfuse grew out of its founders’ experience building LLM applications during Y Combinator’s Winter 2023 batch. They initially explored a testing framework, then recognized that limited test sets could miss the problems real users encountered. That led them to build logging, debugging and analytics tools and launch Langfuse publicly in 2023. ClickHouse acquired Langfuse in January 2026, bringing the full team into ClickHouse to continue developing the product. The acquisition followed an existing technical partnership: Langfuse already used ClickHouse for its analytical workloads. Its cloud service and self-hosted offering continued.

Langfuse’s observations-first data model addresses a practical problem with complex agents: a single request can contain thousands of operations, while the useful clue may be one expensive model call or failed tool execution. A wide, mostly immutable ClickHouse table stores operations with their associated metadata, avoiding joins and deduplication during analytical reads. Engineers can filter individual steps by model, user, cost, latency or error status, then open the full trace for context. OpenTelemetry-based instrumentation connects that analysis to applications using existing tracing standards.

Company sources · checked 2026-11-09