Product Launches and Community Projects

Launch: Langfuse v4 (Langfuse)

Posted by Applied AI Commons team (community team) on

Open-source LLM observability and evals rebuilt for speed and agent-level analysis

Major release of the open-source tracing and evaluation platform: much faster tables and dashboards on large projects, querying and evaluating each LLM call, tool run and agent step, threshold monitors, and Python/TypeScript code evaluators. Live on Cloud and GA for self-hosting. Cloud goes v4-only on Nov 16, 2026, so some projects must migrate.

Released 2026-08-17 by Langfuse. Listed by the community team; we are not affiliated.

How are you evaluating agent runs today: per step, per trace, or only on the final answer?

Link: https://langfuse.com/changelog/2026-08-17-langfuse-v4

3 replies

  1. Applied AI Commons team (starter reply from the community team),

    Per-step evaluation is the important change for agent teams: a final answer can look fine while an intermediate tool call quietly did the wrong thing.

  2. Applied AI Commons team (starter reply from the community team),

    If you use Langfuse Cloud, note the date: Cloud goes v4-only on 16 November 2026, so check whether your project needs migrating before then.

  3. Applied AI Commons team (starter reply from the community team),

    Code evaluators in Python and TypeScript make it practical to write checks such as 'every claim has a source' or 'the SQL query is read-only' and run them on every trace.

More in Product Launches and Community Projects