Your AI assistant has amnesia every single day. Why are we okay with this?
You open Claude or ChatGPT every morning and start from zero. Re-explain your project. Your role. Your context. Again.
We've built incredible AI - and given it no memory of you.
Minimi is our attempt to fix that. It runs quietly on your Mac, reads what you're working on (locally, never uploaded), and builds a personal memory layer that connects to Claude. No manual journaling. No copy-pasting context. It just already knows.
A few things we've learned building this:
Most people don't want to actively "log" anything - ambient capture or nothing
Privacy feels scary until people realise it's all local, then they love it
The highest-value memory is surprisingly mundane: meeting decisions, half-finished tasks, names you saw once
What's your mental model for ambient AI memory? Would you want your AI to remember everything - or would that feel like surveillance?
Replies
To be honest I would rather not want my AI to remember everything, because I consider this counterproductive - for instance, I, as a human, tend to forget the parts of a given system I haven't worked with for quite some time, therefore when I get back to these parts, I need to spend a bit of time on recalling what this is all about. Therefore I think that the AI should work in the exactly same way, because it sounds natural to me
the idea sounds great but remembering everything feels risky. the part I keep coming back to is stale context. old decisions, outdated preferences, projects you've moved on from. if the memory layer doesn't know what's current vs what's obsolete, it can confidently mislead you with yesterday's truth. that's almost worse than no memory at all.
the question isn't how much it remembers, but how well it handles updates. does a new decision override the old one, or do they sit side by side creating noise? that conflict resolution problem feels like the hardest part of building this well.