Langfuse

Recently shipped Home as a customizable dashboard and evaluator tool call support.

Home is now a dashboard you can customize — choose which dashboard shows on Home, reuse Home widgets, and copy, duplicate, and share widgets as portable JSON.

Evaluator tool calls and boolean score filters — access recorded tool calls directly in code and LLM-as-a-Judge evaluators; filter tables by true or false score values.

Project notification channels — send blob export and evaluator deactivation alerts to Slack or a webhook, configurable per project in project settings.

Parquet exports to blob storage — scheduled blob exports now default to Apache Parquet, configurable via the public API.

Experiments public API and MCP tools — fetch experiments and their items programmatically; the MCP server added experiment tools for AI agents.

Dark theme revamp — improved surface elevation, text hierarchy, and semantic status colors across the web UI.

June's launch week brought monitors and alerts, the Langfuse Assistant public beta, a typed filter search bar with AI-drafted queries, and multi-modal dataset support. The MCP server expanded to fifteen tool categories, and Web Callouts and browser-side scoring extended the integration surface.

Across the SDKs: Python SDK v4.14.0 added prompt linking via propagate_attributes and captured service_tier; v4.13.0 reinitializes consumer threads after os.fork(). JS SDK v5.9.1 fixed CJS route resolution, and earlier versions added AI SDK v7 integration and media upload toggles.