Feature Request: Export Human Annotation Comments from Human Annotation Page

[DelJaguar]

Describe the feature or potential improvement

Currently, comments can be added to traces, observations, sessions, and prompts in Langfuse, and the Comments API allows programmatic access to comments. However, there is no documented way to export all comments specifically associated with human annotations or annotation queues directly from the Human Annotation page or through bulk export features in the UI

Feature Request:

Please add the ability to export all comments related to human annotations (including annotation queues) directly from the Human Annotation page in the UI. Ideally, this export should support common formats like CSV.

Use Case:

Enables teams to review, analyze, or archive annotation feedback and discussions offline.
Facilitates audit trails and reporting for manual evaluation processes.

Documents referenced before making request:

Human Annotation documentation
Comments on objects
Export Data from UI

Additional information

No response

[jannikmaierhoefer]

+1

[jannikmaierhoefer]

Thanks for adding this @DelJaguar, and I agree that this would be very helpful! I created a python script to allow you to query the comments and export them as CSV in the meantime: https://colab.research.google.com/drive/1ErETZNWHyOjkG262bZHh-j3Vo9z83qUj

[marliessophie]

I see, thanks for sharing this feedback! Please upvote and let us know should this functionality continue to prove important. In the meantime the script @jannikmaierhoefer created (thanks!) should work well.

[pavzagor]

The workflow I have: I'm looking through the traces to look for anomalies, I want to mark traces somehow to later understand what anomalies I found and to communicate it to my team.

Currently I can use annotate for that but i have to create scores which is not what I need + I can't have a score with multiple values selected.

I want to be able to create tags or something similar on the fly, write comments and then search using those.

[marliessophie]

Tagging @FroeMic who is starting some research in this direction at the moment.