Concepts

This section covers the ideas behind Omnigraph's design. You don't need to read all of it before building, but it helps explain why the system works the way it does.

Knowledge graphs

A knowledge graph gives your agents a connected, queryable world model instead of scattered key-value lookups. Entities have types, relationships have meaning, and traversal replaces joins.

Graph modeling

Graph modeling is the practice of turning your domain into nodes, edges, and properties. The choices you make here determine how naturally your queries read and how well your graph scales.

Ontologies

An ontology is the typed schema that gives your graph a formal domain model. In Omnigraph, the .pg schema file is the ontology -- it defines what exists, what connects to what, and what constraints apply.

Traversal

Traversal is how you move through connected data. Instead of joining tables, you follow edges -- one hop, two hops, or variable depth -- to answer questions about relationships.

Context

Omnigraph works as a shared context store for agents. The schema tells agents what to look for, queries scope what they read, and branches isolate what they write.

Coordination

Coordination is how multiple agents work together through shared graph state. Branches provide isolation, hooks trigger reactions, and queries define the contracts between agents.

Governance

Governance keeps a multi-agent knowledge base trustworthy. Branching, audit trails, and merge gates let you control what changes land and trace every decision back to its source.

Hooks

Hooks let you react to graph changes or query-result changes without building a separate event layer. They are the mechanism that turns a passive data store into an active coordination surface.