Changelog

We've rebuilt Langfuse's infrastructure to address performance bottlenecks at scale.

Capture improved versions of LLM outputs directly in trace views. Build fine-tuning datasets and drive continuous improvement with domain expert feedback.

Anchor comments to specific text selections within trace and observation input, output, and metadata fields.

Add filtering, table columns, and dashboard widgets for analyzing tool usage in your LLM applications.

New high-performance v2 APIs for metrics and observations with cursor-based pagination, selective field retrieval, and optimized data architecture.

Track dataset changes over time with automatic versioning on every addition, update, or deletion of dataset items.

Langfuse now supports OpenAI GPT-5.2 with day 1 support across all major features.

Select multiple observations from the observations table and add them to a new or existing dataset with flexible field mapping.

Langfuse now supports pricing tiers for models with context-dependent pricing, enabling accurate cost calculation for models with context-dependent pricing.

Langfuse now includes a native Model Context Protocol (MCP) server with write capabilities, enabling AI agents to fetch and update prompts directly.

Langfuse now supports OpenAI GPT-5.1 with day 1 support for the LLM playground, LLM-as-a-judge evaluations, and comprehensive cost tracking.

Use slashes in dataset names to create folders for better organization.

Define JSON schemas for your dataset inputs and expected outputs to ensure data quality and consistency across your test datasets.

Validate evaluation reliability and uncover insights with comprehensive score analysis. Compare different evaluation methods, track trends over time, and measure agreement between human annotators and LLM judges.

Add human annotations while reviewing experiment results side-by-side. Review experiment outputs, assign scores, and leave comments while viewing full experiment context.

Compare experiment runs side-by-side with baseline designation to systematically identify regressions and improvements

Filter experiment results in the compare view to focus on specific subsets, such as items where evaluator scores dropped below a threshold

Trace agents with beautifully rendered tool calls and understand their performance through Agent Evals.

Trace AI agents built with Amazon Bedrock AgentCore via OpenTelemetry and Langfuse

Tag teammates with @mentions to notify them instantly, and add emoji reactions to comments for quick acknowledgments.

Langfuse now supports IdP-initiated SSO, allowing users to start authentication directly from their identity provider (e.g., Okta, Azure AD, Keycloak, JumpCloud)

We teamed up with Mixpanel to integrate LLM-related product metrics into your existing Mixpanel dashboards.

The traces endpoint in public API now supports complex JSON-based filtering.

Quickly filter tables by column values in the filter sidebar with one click.

Langfuse SDKs now support Langchain v1 for both Python and JS/TS. The integration remains stable and backward compatible.

Every LLM-as-a-Judge evaluator execution now creates a trace, allowing you to inspect the exact prompts, responses, and token usage for each evaluation.

Monitor your organization's cloud spending and receive notifications when costs exceed predefined monetary thresholds.

Filter your traces using plain English queries. Powered by AWS Bedrock with zero data retention.

Enforce JSON schema response formats in prompt experiments to ensure consistent, parseable outputs for evaluation and analysis.

Score configurations can now be updated after creation. Modify existing configs via API/SDK and UI while keeping all your data safe.

New high-level SDK abstraction for running experiments on datasets with automatic tracing, concurrent execution, and flexible evaluation.

The new OpenTelemetry-based TypeScript SDK v4 is now generally available with improved DX, modular packages, and seamless integrations.

New observation types including Agent, Tool, Chain, Retriever, Evaluator, Embedding, and Guardrail provide semantic meaning to your traces.

We've released new comprehensive walkthrough videos covering observability, prompt management, and evaluation to help you get up to speed quickly with Langfuse.

Find dataset items by searching through their actual content with our new full-text search capability

Set additional provider options in your LLM calls in playground and in llm-as-a-judge evaluations.

Append .md to any docs URL to fetch the page as Markdown. Built at compile time for fast, reliable access.

Day 1 support for OpenAI GPT-5 including tracking token counts and USD spend.

Assign users to annotation queues to make it easier for team members to focus on relevant tasks.

Receive prompt change notifications directly in your Slack channels with our native integration.

The LLM Playground now supports side-by-side prompt comparison with parallel LLM execution.

Annotation Queues now support session-level annotation, making it easier to evaluate multi-turn interactions in your LLM applications.

Trace real-time voice AI agents and multimodal conversations built with LiveKit Agents via OpenTelemetry and Langfuse

Enable non-technical users to trigger external experiment pipelines directly from the Langfuse UI on datasets.

Find prompts by searching through their actual content with our new full-text search capability

Self-hosted deployments can now use the AWS SDK default credential provider chain for Amazon Bedrock LLM connections.

Get real-time notifications for new prompt versions, changes to deployment labels or tags, and other edits.

Use Langfuse's prompt management capabilities directly in your n8n workflows to build powerful no-code agents.

Langfuse says hello to Gemini 2.5 Pro (GA), Gemini 2.5 Flash (GA), and Gemini 2.5 Flash Lite (Preview). Support added for the LLM playground, evaluations, and cost tracking.

Inject context or message history into your chat prompts at just the right spot.