
Langfuse May Update
Langfuse May Update: Python SDK v3 GA, Launch Week #3 Recap, Custom Dashboards & More!
May 2025 was our most action-packed month yet at Langfuse. We welcomed thousands of new users, powered through countless lines of code, wrapped our third Launch Week, shipped long-awaited API goodies, and rolled out a handful of fresh integrations.
🔭 Python SDK v3 is now Generally Available
Following extensive testing and community feedback, we’re excited to announce that Langfuse Python SDK v3 is now generally available and ready for production use.
Why you’ll love it:
- Unified context propagation: Automatically nests spans and generations, even across threads or async tasks.
- Simple & Flexible Instrumentation: Use the
@observe
decorator, context managers, or manual observations. Mix and match as you like. - Ecosystem Integration: Works out-of-the-box with OTel-instrumented libraries like MLflow, OpenLit, and more – covering the long-tail of model and agent SDKs much better.
- Global Client Access: Easily fetch the initialized Langfuse client anywhere in your project via
get_client()
. - 🕹️ Get started now:
pip install langfuse
→ Find out more in the migration guide & docs
⭐ Launch Week #3 Recap
We marked our third Launch Week in May, rolling out six new releases that unlock deeper insights into your LLM data and make it even easier to integrate Langfuse with your favorite tools:
🔎 Day 1: Full-text search lets you query inputs & outputs across traces and observations.
📌 Day 2: Saved table views save your favorite filter combos with a single click.
📊 Day 3: Custom Dashboards let you turn raw data into insights with a wide range of chart types, flexible metrics, and rich filtering options.
📦 Day 4: Hyperscaler Terraform modules for fast and scalable self-hosting.
🔭 Day 5: OpenTelemetry Python SDK, now GA.
📐 Day 6: Langfuse Evaluator Library for state-of-the-art LLM-as-a-Judge evals.
→ Find out more about all Launch Week releases here.
🔍 How we built Customizable Dashboards
In a new blog post, Steffen walks through the product thinking and architecture behind our most-requested feature of Launch Week: Custom Dashboards.
You’ll see how a virtual data model decouples the UI from ClickHouse, how the query builder assembles complex charts in milliseconds, and why reusable widgets make collaboration a breeze.
→ If you’re curious about the trade-offs that let dashboards stay flexible and fast, even on billions of rows, give the post a read.
➕ More May Releases
- Metrics API: Query custom dimensions & time-series straight from your data warehouse or BI tool.
- Pipecat tracing integration: Real-time audio agents now send spans to Langfuse via OTel out of the box.
- Claude Sonnet 4 & Opus 4 support: Day-1 playground, evaluator, and cost-tracking support.
- Customize product-module visibility: Hide or whitelist sidebar modules with two env vars when self-hosting Langfuse.
- Dataset run-level scores: Record aggregate precision/recall/F1 (and more) for every experiment run.