Langfuse

Fixed OTel int64 field serialization to use strings instead of protobufjs Long internals. Monitors now support metadata filters, and a new NEXTAUTH_COOKIE_NAME_SUFFIX option namespaces auth cookies across instances.

Projects using app-root-aware Langfuse SDKs now open the observations table with a focused view of outer roots and application entry points by default.

Monitors are now available for self-hosted deployments with events writes, and the code evaluator editor includes contract-aware autocomplete. The worker no longer kills background-migration chunk queries from the 35s max_execution_time, and eval preview tables honor the date-range window.

Trace graph view now offers two layouts: an aggregated overview and an expanded as-it-ran DAG, backed by a rebuilt deterministic renderer.

Added support for prompt linking via propagate_attributes in tracing, and OpenAI service_tier is now captured in model_parameters. Fixed handling of Anthropic cache_creation nested dict in usage reporting.

Langfuse tables now support filtering by true or false score values.

Recorded tool calls can now be accessed directly in code and LLM-as-a-Judge evaluators.

Home is now a customizable dashboard where you can choose which dashboard shows on it, reuse Home widgets in your own dashboards, and copy, duplicate, and share widgets as portable JSON.

The in-app agent now displays live reasoning, and the Home dashboard can now be customized with preset widgets and per-project selection. Added a graph view mode switcher for traces (aggregated vs expanded DAG), PostHog instrumentation skill for agents, and cloud admin access to org and project settings pages.