I made a memory cleaner for Claude. I guess it'll probably work in other AIs but I've only tested it in Claude and it's for getting rid of all of the bloat and crap that builds up in your memory file so that it works better because when it's too big it stops working.

I built it because I noticed that over time my Claude sessions were missing some of the things that I'd asked it to remember and it was full of contradiction.

The reason why this works is because it always asks you for your opinion rather than just going off and doing a prune on its own. If you don't trust the model to not bloat your memory in the first place you shouldn't trust it to be the one who prunes that memory either. That's why it has to be an interview-like, diff process. And guess what? It works because you do it that way.

Anyone else noticed theirs bloat? Found anything daft still lurking in there? Mine was full of anecdotes :D