Source: github.com/langfuse/langfuse/releases/tag/v4.78.0
Langfuse v4.78.0 was released on 10 November. On the AI Gateway side, custom models can now be offered everywhere with a chosen connection, and Bedrock is available in every environment. The AI Gateway navigation is always shown at organization level and linked from projects.
Evals gained an external HTTP code eval dispatcher. Topics setup can now filter traces, and each rule's map is refit nightly as a continuous map, while manual runs are kept as snapshots. Prompt cache outcomes are now tagged on request spans.
Access control changed as well. A USER API key scope is evaluated against user and key policies, and memberships are mirrored into role assignments. Fixes include sending max_completion_tokens to OpenAI connections with custom base URLs and rejecting unqualified columns in SELECTs with several relations.
Why it matters if you build agents
If you route agent model calls through Langfuse's AI Gateway, custom models can now be set up in every environment with a connection of your choice. Teams that score agent outputs can call an external HTTP service for code evals.
If you issue user-scoped API keys, check how the new USER scope interacts with your existing key and user policies before relying on it.
Get stories like this every weekday
One short email with the day's agent tooling news.