v2.3 shipped today with tiered degradation, veracity signals, smart compression, and cross-provider importers. The docs are live with honest comparisons against every major memory provider. I've been thinking a lot about what comes next.
Hermes-first, but not Hermes-only
Mnemosyne was built to be the memory system Hermes Agent should have had from day one. That hasn't changed. It's still Hermes-first. Every API decision, every optimization, every tool integration starts from the question: does this make Hermes better?
But here's the thing. I've been watching people use Mnemosyne in ways I never planned. Wysie built a dashboard on top of it. People are running it standalone, outside of any agent framework. It's becoming clear that Mnemosyne can be more than a plugin. It can be a general memory framework that benefits anyone running agents, CLI tools, terminal workflows, anything that needs persistent memory.
So that's where my head is at. Push Mnemosyne beyond the Hermes ecosystem while keeping Hermes the first-class citizen it's always been. Make it so anyone can pip install it and have a working memory layer in five minutes, regardless of what framework they use.
The cloud question
Someone in the community, steezkelly, brought up the idea of a hosted version. I'll be honest: this one makes me pause. Mnemosyne's inception was built on two principles: local-first, privacy-focused. Your memory lives on your machine. No external API calls unless you opt into them. No vendor lock-in. That's been the line since day zero and it's not something I take lightly.
A cloud version could take away from those principles. I don't want Mnemosyne to become another SaaS memory platform where your data lives on someone else's server and you need a subscription to access it. That's exactly the problem I was trying to solve.
But I also get it. Not everyone wants to manage a SQLite database. Not everyone has persistent storage on their deployment. Some people just want to sign up and go. There's probably a middle ground somewhere: bring-your-own-storage, or an optional sync layer that keeps local data local while enabling cloud convenience for those who want it.
I haven't figured out exactly what that looks like yet. What I do know is that the local-first, privacy-first principles are not negotiable. Whatever comes next, Mnemosyne will never force you to send your data anywhere you don't want to.
Where this is going
The immediate roadmap is straightforward: keep shipping solid features, keep the docs honest, keep the comparisons accurate, keep the community growing. The long game is bigger. Make Mnemosyne the memory layer that works everywhere, for everyone, on your terms.
That's the vision. Let's see where it goes.