Changelog

Improved In-Product Onboarding

New onboarding screens that introduce the capabilities of Langfuse features that are not yet used within a project.

Claude 3.7 Sonnet support

Langfuse says hello to Anthropic Claude 3.7 Sonnet with day 1 support for both the LLM playground, LLM-as-a-judge and cost tracking.

Google AI Studio support

Langfuse now supports access to Gemini models via Google AI Studio.

Execute LLM-as-a-judge evaluations on existing data

You can now choose to execute LLM-as-a-judge evaluations on existing, new, or all data.

Model Context Protocol Prompt Server

Use Langfuse prompts in your LLM Agent systems, Claude Desktop, and Cursor and other MCP Clients via the new MCP Server.

OpenTelemetry Tracing Support

Push your OpenTelemetry Spans to Langfuse

Open Source LLMOps Stack

Together with LiteLLM, we launch the OSS LLMOps Stack which is a popular, well-integrated and battle-tested LLMOps stack that can easily be self-hosted and extended.

Graph view for LangGraph traces

Follow your LangGraph agent execution on a graph view

Select all items in tables with a single click

Quickly select all items subject to your filters and perform actions on them.

Trace UI filtered view by log-level

Focus on the most relevant elements of your trace by using log-level filters

Gemini 2.0 support for playground and cost tracking

Langfuse says hello to Gemini 2.0 Flash, Gemini 2.0 Flash-Lite, and Gemini 2.0 Pro. Support added for both the LLM playground and cost tracking.

Use JsonPath to select from Input, Output, or Metadata values in LLM as a Judge

Use JsonPath to select from Input, Output, or Metadata in LLM as a Judge for more precise evaluation prompts.

Typed public API client in SDKs

Access all of Langfuse public API resources via a typed client in our SDKs

OpenAI o3-mini support for playground and cost tracking

Two hours ago, OpenAI released the latest version of their o3-mini model. Langfuse now supports this model in both the LLM playground and cost tracking.

Cmd+K Menu

Quickly access different Langfuse features and switch between projects via the Cmd+K menu.

Project-level Data Retention Policies

Automatically remove event data outside of a specified retention period.

New API Reference

Langfuse now has a new API Reference with interactive examples.

New Prompt Editor

Complex prompts are easier to edit with our new prompt editor, which supports variable highlighting.

JS/TS SDK supports trace sampling

Trace sampling is now supported in the JS SDK and integrations.

Commit Messages on Prompt Versions

Document your updates to prompts

Dataset CSV upload

Create your Datasets in seconds

Configure additional HTTP headers for LLM API keys

Configure additional HTTP headers for LLM API keys to include custom headers in requests to provider endpoints, particularly useful for OpenAI-compliant LLM proxies.

Track changes between prompt versions

See a detailed comparison of changes between prompt versions. Optionally, you can also review changes before creating a new prompt version.

Audit Logs

As part of our ongoing effort to improve Langfuse for larger teams, we are excited to announce that audit logs are now available on Enterprise/Teams plans.

PostHog integration is now GA

After running for some time in public beta, we are excited to announce that the PostHog integration is now fully available. It runs on a hourly basis and is available to all Langfuse plans (cloud) and Pro/Enterprise (self-hosted).

Improved cost tracking

Langfuse now supports cost tracking all usage types such as cached tokens, audio tokens, reasoning tokens, etc.

Langfuse v3 stable release

Langfuse v3 is now stable and ready for production use when self-hosting Langfuse, including many scalability and architectural improvements.

Extensive example notebook for JS/TS SDK

To make it easier to get started with the Langfuse JS/TS SDK, we've created an extensive end-to-end example notebook. It includes a general introduction and examples for Anthropic, OpenAI SDK and LangChain.

SSO with GitHub Enterprise and Keycloak

Langfuse now supports GitHub Enterprise and Keycloak as SSO providers on Langfuse Cloud and self-hosted instances

New documentation for Google Vertex AI and Gemini tracing

Comprehensive guides for tracing Google Vertex AI and Gemini models with Langfuse

Google Vertex AI support for LLM Playground and Evaluations incl. Gemini models

Langfuse now supports Google Vertex AI incl. Gemini models for LLM Playground and Evaluations.

Launch Week 2 🚀

Prompt Experiments on Datasets with LLM-as-a-Judge Evaluations

Move fast on prompts without breaking things! Run experiments on Datasets and directly compare evaluation results side-by-side. Experimentation speeds up the feedback loop when working on prompts and prevents regressions when making rapid changes.

Launch Week 2 🚀

All new Datasets, Experimentation and Evaluation documentation

We've completely rebuilt the documentation for Datasets and Evals to make it easier to get started with offline evaluation. To celebrate Launch Week, we've also summarized all the documentation improvements we've made over the past year.

Launch Week 2 🚀

Full multi-modal support, including audio, images, and attachments

Add multi-modal attachments to LLM traces in Langfuse to observe, and view them within the Langfuse UI.

Launch Week 2 🚀

LLM-as-a-judge Evaluators for Dataset Experiments

Introducing support for managed LLM-as-a-judge evaluators for dataset experiments.

Launch Week 2 🚀

Dataset Run Comparison View

After running experiments on datasets, you can now compare results side-by-side, view metrics, and peek into details of each dataset item across runs.

Launch Week 2 🚀

llms.txt

Easily use the Langfuse documentation in Cursor and other LLM editors via the new llms.txt file.

Launch Week 2 🚀

Prompt Management for Vercel AI SDK

Langfuse Prompt Management now integrates natively with the Vercel AI SDK. Version and release prompts in Langfuse, use them via Vercel AI SDK, monitor metrics in Langfuse.

New Sidebar

We've updated the sidebar in the Langfuse UI collapsible. This change provides more screen space for viewing traces and other complex pages.

Event input and output masking

Configure SDK-side masking to redact sensitive information from inputs and outputs sent to the Langfuse server.

Amazon Bedrock support for LLM Playground and Evaluations

Langfuse now supports Amazon Bedrock for LLM Playground and Evaluations.

Langfuse LLM-as-a-judge now supports any (tool-calling) LLM

Tool calling makes Langfuse Evals reliable. Previously, only OpenAI models were supported. With this update, you can use any tool-calling LLM when setting up an LLM-as-a-judge evaluator.

Annotation Queues

Manage your annotation tasks with ease using our new workflow tooling. Create queues, add traces to them, and get a simple UI to review and label LLM application traces in Langfuse.

Aggregated and Color-coded Latency and Costs on Traces

Large traces can be hard to read. We've added aggregated latency and cost information to the every span level to make it easier to spot outliers and debug the LLM application.

Documentation now integrates with GitHub Discussions (Support and Feature Requests)

See previews of all commonly asked questions and feature requests for each page across the Langfuse documentation.

Langfuse on AWS Marketplace

Langfuse is now available on AWS Marketplace, making it the premier choice for AWS customers looking for a powerful LLM observability, analytics and evaluation platform.

DSPy Integration Example

Langfuse now provides an integration example for DSPy, offering observability for this powerful framework that optimizes language model prompts and weights.

Link prompts to Langchain executions

Prompt management just got more powerful for Langchain users by linking Langfuse prompts to Langchain executions.

Custom Base Path for Self-Hosted Deployments

Deploy Langfuse behind a custom base path for more flexible self-hosting configurations.

Token/cost tracking for OpenAI's o1 models

Day 1 support for OpenAI's o1 models including tracking token counts and USD spend.