Auto categorize incoming mail
Is your feature request related to a problem? Please describe.
The inbox is busy with messages and things are hard to find
Describe the solution you'd like
Automatically put messages in categorized buckets
https://github.com/user-attachments/assets/b7b89773-070b-4a39-abf6-16a3805510c9
Describe alternatives you've considered
No response
Additional context
No response
@ChristophWurst this is related to #11062 but not quite the same
As I understand it focused and other are just two categories, important and not, which also overlaps a bit with priority mail I would say. These you see in the video would be several buckets with specific categories. "Focused" could be what in the video is shown as "primary"
Okay, makes sense. The topic of categorization was raised in the past. The challenge is that we don't have any data to train a model, and there was none available off the shelf for this purpose.
I'll tag this as accepted with no delivery promises, while the technical feasibility is still to be clarified.
For reference, GMail has a similar feature since quite some time already:
Not entirely sure if this is the default or if I ever changed something:
Yep. I think the feature is clear. The open question is how an open source software like Mail can achieve the same. We can throw an LLM at it, but I think it's really inefficient. My best guess is that big tech buys or builds labeled training data to train traditional ML.
Technical feedback I received from @marcelklehr is that this will be feasible with our already existing freeprompt task type. It will however be quite slow on selfhosted models (given that some people have a huge amount of emails). But with ChatGPT it could be fast. Another option could maybe be to use the vectordb of context chat but we didn't think that part through yet.
Some ideas:
- Allow users to configure their own categories
- Allow users to turn the feature off (to reduce load on the system for those who don't want the feature)
- Only enable for "new" emails to reduce load of the system when facing large inboxes
If it is clear when the groupware team would like to work on this we can assist with the prompt design.
Thanks for the input @DaphneMuller! We discussed piping all email through an LLM for this purpose and considered it too expensive in terms of resources needed, while it's technically possible. The feature is not planned at the moment.