Memory that AI agents can walk
A persistent, connected memory an AI agent traverses, instead of a context window it forgets between turns.
AI Research
We believe the future of AI is graph-shaped.
Language models reason brilliantly and remember unreliably. We think the missing piece is structure: knowledge shaped as a graph of what connects to what, so a model can be grounded in something durable and inspectable.
The thesis
Models reason. Graphs remember.
Bigger models keep getting smarter, and that alone will not fix what a model does not know about your world. The gains left on the table are in the context: the right, connected, permission-aware knowledge delivered the moment a question is asked. A graph of pages, concepts, and policies is a memory a model can be held to.
Where it ships
Our retrieval engine is the research, in production.
Everything we learn about graph-shaped context goes straight into GraphStaff’s graph retrieval engine: how to model relationships, how to walk them, and how to assemble cited, traceable, permission-aware context before a model answers. The product is where an idea has to survive contact with real questions.
Open questions
We are following the graph past retrieval.
Graph retrieval is the first place these ideas ship, not the last. A few of the questions we are working on now.
Retrieval that reasons over relationships
Following dependencies, prerequisites, and exceptions the way a person would, not matching text and hoping.
Grounding you can verify
Measuring whether an answer truly came from the source it cites, so trust is earned instead of assumed.
A long-term bet This is where we’re
This is where we’re
placing it.
Graph-shaped knowledge is the throughline from our research to the answer your customers and their AI get today. If that is a bet you want to see up close, come talk to us.