- Architecture
- Graph-native
- Algorithm
- Spreading activation
- Scope
- Cognitive infrastructure
Problem
Personal and project knowledge lives in fragments — notes, chats, commits, documents, memories. Traditional search treats these as independent lookups. Real recall doesn't work that way: one thought activates another, context pulls context. We wanted a system that behaved more like how thinking actually works.
Approach
01
Graph-native data model. Nodes for concepts, people, projects, events; edges for relationships, references, and co-occurrence. Context is first-class, not derived.
02
Spreading activation retrieval. A query activates not just direct matches but the neighborhood around them — surfacing what's connected, not just what's nearest in embedding space.
03
Go-based monorepo with a small dependency surface. The substrate is supposed to outlive the frameworks it's built on.
04
Designed as infrastructure, not an application. Other tools sit on top and borrow its memory.
Outcome
A working cognitive substrate that informs how we think about memory, context, and knowledge systems for every other engagement. What we learned here shows up in how we architect regulated AI elsewhere — graph-first, context-aware, built to be queried the way people actually think.
Stack
- Go
- Graph databases
- Spreading activation
- Monorepo architecture
- Cognitive infrastructure