Building the memory layer for AI that remembers
How AI agents and copilots hold onto context is at an inflection point. We build the infrastructure that lets them remember, on hardware you control.
So what do you actually do?
You'd think this would be a short answer, but the honest version has two halves.
Build a memory engine? Yes.
Research how memory should actually work in AI systems? Also yes.
Esment is our only product — a local-first memory engine that gives AI agents a temporal knowledge graph, a git-like commit history, and a retrieval pipeline that runs entirely on hardware you control.
---
Why does NotasAI exist?
We believe AI systems deserve persistent, structured memory — not just a bigger context window.
Alongside the product we run an ongoing research effort into how large language models remember: what gets forgotten, what gets reinforced, and how a knowledge graph should decay and consolidate over time.
Those findings ship first inside Esment — the lifecycle engine, the importance-scoring formula, and the retrieval cascade are all directly informed by it.
---
How do I get in touch?
Want to try Esment, talk about a licensing deal, or compare notes on memory research?
Either way, we'd like to hear from you.