CogitaveLearn
Training
View as
MarkdownJSON

Introduction

Ask five different systems where a piece of company knowledge lives - the docs site, the internal developer portal, the compliance evidence store, the infra inventory, an agent's retrieval index - and a normal company gives you five different answers, in five different stores, that drift from each other the moment two of them are updated separately.

An AI-native organization refuses that split. It runs on a single typed property graph that all of those surfaces are projections of - not synced copies, not separate stores that happen to agree today. Docs, the internal developer portal, governance and compliance evidence, infrastructure inventory, and agent knowledge all read from the same graph, through the same query layer, and that layer is MCP-native: a human using the API/UI and an agent using MCP get the same knowledge from the same model. Cogitave's Core is one such graph - the worked example this module builds on. Its canonical architecture doc states the same thesis for that instance, and this module - and the rest of this path - builds directly on it.

What this module teaches

Two ideas make the one-graph claim more than a slogan:

  • The one-model thesis - why one graph, not one store per domain, and what that buys in exchange for the discipline it demands.
  • The property-graph substrate - why that one graph is specifically a labeled property graph, what a design decision like Cogitave's ADR-0001 accepted and rejected to get there, and what makes it fast enough to serve.

Status: read this before you build against it

Cogitave's architecture doc opens by calling itself a "Day 0 canonical spec." Treat it that way here, too. A one-graph model and the decision behind it are things you decide - accept, not merely propose - and this module teaches Cogitave's as documented decisions. It is not a claim that every product has finished projecting into the graph, and it is not a hands-on module: you will not stand up a canonical model here, you will learn to read the spec that governs one, the same spec an auditor or a querying agent relies on.