The graph is the memory.
Sessions are ephemeral.
Substrate memory for AI — the typed graph parallel sessions reason on, so they stay coherent without a human carrying decisions between them.
A substrate,
not a scaffold.
A scaffold wraps a session in templates and tells it what to do next. Vinculum does the opposite — it provides a persistent, shared, typed medium your sessions use however the work demands.
The intelligence stays with the model. The substrate holds what the model can't: everything decided before this turn.
Claude writes entries into the graph; other sessions read them; the dashboard renders every write in real time. You're not reading a log — you're watching the substrate grow.
It started as
“Netdata for AI.”
I was running four, six, sometimes more parallel Claude sessions on real software, and I couldn't see what any of them were doing without tabbing between windows reading walls of output. I wanted a dashboard. Netdata, but for my AI sessions — just show me what each one is working on.
So I built the eyes. And building the eyes made it obvious: the sessions didn't just need to be watched — they needed to see each other. The dashboard needed something to display, which meant the sessions needed somewhere to write, which meant a shared place to read from.
The monitor needed a memory underneath it. And the memory turned out to be the whole thing.
The dashboard is still there. It's just no longer the point — it's the window into the point.
A project returned to over days and weeks has no memory of itself — every session re-derives what's already been figured out. And the moment you run more than one session, you also become the coordination layer, hand-carrying decisions between windows. Vinculum routes both through the graph: the project remembers itself, and sessions read what sessions wrote.
“The model can hold a conversation, not a project. Every session in a long-running project pays the same tax: re-derive what's already been figured out. The substrate is what makes a project remember itself.”
AGPL v3. No moat in the code.
Open source under AGPL v3, commercial license available for orgs that can't comply with copyleft. Self-host on your own hardware, or run hosted at vinculum.run — same code, managed ops. The hosted tier exists for people who don't want to run a box, not as a wall around features.
No Kubernetes. No microservices. No cold starts. No VC runway.
Latin for bond, link, that which binds. In mathematics, the bar over a repeating decimal — the mark that says these digits recur, indefinitely, as a unit. That's the architectural claim: a substrate that accumulates a project's intent across its whole life, holding parallel sessions together not by syncing them, but by giving them one authoritative surface to read from and write to. The sessions are ephemeral. The graph is not.
— the substrate that persists —