Langfuse

Recently shipped Langfuse v4 GA, unlocking full-text search, monitors and alerts, and faster Observations and Metrics API v2 for self-hosted deployments.

Langfuse v4 reached general availability — The stable v4.0.0 release brings full-text search across inputs, outputs, and metadata, a filter search bar, and significantly faster Observations API v2 and Metrics API v2 to self-hosted setups. The v3 maintenance line is winding down — its latest markers were retired — and a dedicated guide covers the v3-to-v4 migration, which drops superseded Postgres and ClickHouse tables.

Trace and session views refined — Mobile tabbed trace view (Tree/Timeline/Graph/Info), a span list in the modern session sidebar, and media size hints in previews landed across v4 point releases. The virtualized lazy JSON renderer remains the viewer for large traces.

Ingestion overflow moved to media — Oversized direct-event fields and observation field overflows now spill to media instead of being dropped, and the worker skips token counts rather than falling back to synchronous tokenization.

In-app agents gained background execution — Long-running in-app agents now run in the background under a canonical hydration contract; drafts stay editable while running and the maximum agent step count increased.

Metrics semantic roots surfaced — The v2 API exposes semantic roots, and dashboards forward semantic root filters to v4 tables.

SDK updates: JS v5.10.0 deprecated the link method in favor of runExperiment() and added prompt linking via propagateAttributes. Python v4.14.2 normalizes LangChain tool definitions to OpenAI function format.