A better kind of memory for your AI Agents

Everyone is excited about Claude + Obsidian right now.
It feels like progress. You keep notes in one place and feed them to your AI. It's convenient.
But it's not memory.
Real memory doesn't just store what you wrote. It remembers for you across every AI you use. It understands what matters today versus what mattered last year. It connects ideas by meaning instead of keywords.
And it protects what's private without you babysitting every prompt.
Obsidian + Claude is still just a smarter notebook plus context stuffing.
You end up repeating yourself. You leak details you didn't mean to share. And you pay token taxes to re-read entire documents every single time. The memory still lives half in the cloud vendor's world, not fully yours.

That's why I built something different.

It's a private, local memory layer you own completely.
One encrypted file on your machine. No accounts, no telemetry, no vendor lock-in.
Tell any connected AI once that you drink oat milk, live in Berlin, or that your cat is Pixel, every AI remembers, forever. Memories are linked by meaning, not strings. "What do I drink?" surfaces the right answer even if the wording has changed. "Where do I live today?" ignores last year's Lisbon entry. A local firewall sits in front of every AI call. It redacts sensitive information, respects your private flags, and gives you a signed receipt of exactly what was shared. You can watch live which memories are being touched.
For documents, it uses structured understanding (we call it SemGraph).
Drop in a 50-page note. The AI reads a cheap outline first, then only the relevant section. Same quality answers at roughly 29% of the tokens.
You can open the memory graph, explore it, edit it, delete anything or ask AI agent to do so.

The black box, turned glass.

This isn't another note-taking hack. It's the memory layer your AIs have been missing: private, semantic, cheap to read, and fully under your control. If you're tired of renting fragments of context from different providers and want memory that actually feels like yours, I'd love to show you what we've built.

🚀Launching soon

Early access list will be opened soon for those who want the real thing.