Composable prompts -> linking traces/evals with prompt fragments

[ozerum]

Describe your question

I am new with Langfuse and evals so apologies in advance if this is an obvious issue. I have read docs, talked to Langfuse AI and searched through discussions here and couldn't find an answer.

My use-case:
Multiple prompt fragments, stitched together conditionally depending on the question.
I want to measure the impact of the change of a single prompt fragment to the final result.

A quick example:

• 1. identity.md -> same across all scenarios
• 2. db-A-schema.md -> conditional if question is about that db provider
• 3. db-B-schema.md

If i create a composed prompt that says 1 & 2 or 1 & 2 or 1 & 2 & 3, traces get attributed to that composed prompt v1.

If i change 2 above to v2, composed prompt still stays at v1. since 2 is not attributed to any trace, I cannot measure the performance impact.

Even though I am new, I think this is a very common use-case, and a bit surprised to see it is not solved (not just Langfuse). Maybe I am doing something wrong?

Thank you 🙏

Langfuse Cloud or Self-Hosted?

Self-Hosted

If Self-Hosted

No response

If Langfuse Cloud

No response

SDK and integration versions

No response

Pre-Submission Checklist

• I have checked for existing issues/discussions and consulted Langfuse AI.

[jannikmaierhoefer]

Thanks @ozerum. Linking multiple prompts to one generation is not supported yet but we have this on our roadmap: https://github.com/orgs/langfuse/discussions/7608 Feel free to upvote this post and add any additional thoughts.