Yeah, that’s the part I keep coming back to.
Saving memories sounds simple. Deciding what happens when two memories disagree is where it gets tricky.
I don’t want Haven making those calls quietly in the background. The person using it should be able to see it and fix it.
Been busy, but just wanted to pop in to say, @elevenlabs.io has accepted my application for a startup grant, with a WHOLE Bunch of credits to use with Haven. LETS GOOOOOOO! #Haven #startup
Yeah, I like that way of thinking about it.
For Haven, I don’t want memory to mean “save this forever as truth.” It needs to be more like: here’s the context that seems useful right now, and here’s a way for the person to correct it when it’s wrong.
Still early, but this is the rabbit hole.
This is why AI memory has to be treated like real product infrastructure, not a gimmick.
For Haven, I keep coming back to the same idea: memory should have context, sources, confidence, and user control, especially when the system gets something wrong or memories conflict.
Not at that scale yet, but that’s exactly the problem I’m trying to avoid with Haven. Memory can’t just be “latest write wins.” It needs source/context, confidence, user review, and a way to resolve contradictions instead of quietly rewriting someone.
In early prototype stage… 😏
This is exactly the kind of memory control I’m building toward in Haven.
Not hidden state. Not “trust us.”
A place where users can see what Haven remembers, why it matters, and eventually edit, pin, or remove it.
This is the core challenge I keep coming back to with Haven. Real AI companionship isn’t just better replies — it’s memory that is useful, consent-based, editable, and human enough to understand why something matters.
Exactly. “Memory” can’t just mean saved preferences or a longer chat log. For Haven, the hard part is making memory feel useful, consent-based, editable, and emotionally aware — without pretending it’s magic.