May 2025 Summaries

Langfuse's May 2025 update was a significant milestone, featuring the general availability of Python SDK v3, designed to enhance production use with features like unified context propagation, flexible instrumentation, and seamless ecosystem integration. The third Launch Week introduced six new releases, including full-text search for querying data, saved table views, and customizable dashboards that provide comprehensive data insights through various chart types and metrics. Other highlights included the general availability of the OpenTelemetry Python SDK, Terraform modules for self-hosting, and the Langfuse Evaluator Library for advanced LLM evaluations. The customizable dashboards were particularly noteworthy, offering a flexible and fast solution for analyzing vast amounts of data, thanks to a virtual data model and reusable widgets. Additionally, the update included a new Metrics API, real-time audio agent tracing integration, and support for Claude Sonnet 4 & Opus 4, with options for customizing product-module visibility and recording dataset run-level scores.

Langfuse has introduced customizable dashboards, a much-requested feature that enhances the ability to visualize LLM usage directly within its UI, allowing users to track metrics such as latency trends, user feedback, and cost-performance correlation. These dashboards can be tailored to different user needs, such as product managers or developers, by offering multiple dashboards and a variety of charts and insights. The development process involved creating a flexible query engine, using ClickHouse for efficient query execution, and ensuring the dashboards are decoupled from the data model to allow adaptability. The architecture includes a virtual data model and a custom query builder, which facilitate the creation of flexible and responsive dashboards. The feature was initially launched as a beta on the cloud platform to allow for modifications before its release for self-hosted users, and it received feedback leading to additional features such as resizable widgets and more chart components. Langfuse also plans to offer managed dashboards and additional chart types, with ongoing improvements based on user feedback. The company is actively engaging with the community on platforms like Discord and is hiring for its Berlin office to tackle similar challenges.

Langfuse Launch Week #3, taking place from May 19 to May 24, 2025, unveils new features each day aimed at enhancing LLM applications through deeper insights and seamless integrations. Highlights of the week include a full-text search capability, allowing users to easily locate specific keywords within Langfuse traces, and the ability to save and share table views for streamlined collaboration. Users can also create custom dashboards to turn raw data into valuable insights and utilize new Terraform modules for quick deployment on major cloud platforms like AWS, GCP, and Azure. The introduction of an OTEL-based Python SDK simplifies tracing operations, and the Langfuse Evaluator Library offers improved tools for evaluating LLMs with prebuilt templates and customizable configurations. These updates are designed to make data work more efficiently for users, encouraging collaboration and more effective application management.