Ask HN: Anyone using an LLM to filter email?

1 pointdoctoboggan3 comments

I have a pretty large backlog of unread emails. I was hoping to use an LLM to help me categorize. Most of it is spam ("marketing emails")

I would love it if I could use a local model (I am running ollama locally), but I would settle for a hosted model if needed.

Anyone here have a good experience going down this route?

View on HN

I have not, but I just asked chatGPT and it, chatGPT, suggested a starting point solution.

I asked:

how would I:

- use IMAP to read email from my email server

- call a local LLM (say ollama) to classify the email

- file each email into one of N folders or maildirs or mboxes ?

and received a reasonable python program.

Yeah that's definitely a reasonable idea, thanks

or use POP , POP3, whatever, which will delete the email off the server - which (unpopular opinion) is a feature not a bug.

Yes, one copy of each email; yes, I don't want to handle each (any) email multiple times, even if the second is a once-a-week mass delete of have-processed-already emails.

Reply on news.ycombinator.com