What's this about?
Memory lets agents accumulate experience, stay personalized, and adapt over long interactions. But most memory systems are fixed once built, which limits how much they can improve. MindMemOS proposes a portable memory layer that keeps evolving and can move between agents.
What exactly is agent "memory"?
A model has no memory between conversations — each session starts blank. Agent memory is the machinery bolted on to fix that: storing facts from past sessions and retrieving the relevant ones later. The design decisions are what matter — what's worth keeping, how to find it again, when to update or discard something that's no longer true. Most systems freeze these policies at build time, which is what "self-evolving" here means to change.
Why is this worth noticing?
- Portability is the more unusual claim: memory that isn't tied to one agent implies your accumulated context could follow you between tools rather than being locked to a vendor.
- It's the same theme as SkillZip and the skills economy from a different direction — the durable value is accumulating in the layer around the model, not in the model itself.