Spinning up the flight deck…
Every conversation, idea, and decision worth keeping becomes a memory. Memories link into a graph. The graph lives in a vault you own — private, permanent, portable — and every agent you trust can remember with you. Never lose context again.
Context windows forget. Chat histories belong to platforms. A memory graph in your own vault is the third way: durable like a diary, connected like a mind, portable like a file.
Chats scroll away. Tabs close. Context windows fill up and quietly forget. A memory saved here stays — the idea from March is still linked to the decision it led to in June.
Your memory graph lives in your vault, under your account — not inside a model's weights or a platform's silo. Export it, back it up, walk away with it. That's what self-custodial means.
Any agent you trust can read and write your memories over MCP. Tell one assistant something once, and the next one — on a different model, in a different app — already knows.
Something worth keeping — a fact, a preference, a half-formed idea — saved with one command, from chat or from an agent.
New memories link to old ones with typed edges. Two notes become a relationship; a pile of moments becomes a graph.
Weeks later, you (or your agent) pull the thread and everything connected to it comes back — with the reasoning intact.
The graph grows with you — versioned, revertible, and permanently, unmistakably yours.
Recall by relationship beats recall by keyword. When memories carry typed edges — who, what, because — you don't just find a similar sentence, you follow the reasoning that connected two moments in the first place. And because every change is versioned, you can watch your second brain grow — and rewind it if it picks up something wrong.
Building agentic systems? The same graph has a systems-facing address: memory.delivery — persistent memory, delivered to your agents.