Context Graph

A markdown notes file plus a vector index works for lookups, but fails on real business tasks that span entities, time, systems and decisions. Why retrieval is an attention problem, and when a context graph is worth it.

An AI model settled open questions across seventeen areas of mathematics by reading fields that specialists rarely read together. The same idea, with outcomes as the feedback loop, works on a B2B sales pipeline.

Knowledge-representation researchers split knowledge into a rulebook and a diary forty years ago. Keeping the T-box and the A-box apart is what turns an agent with a long context window into one with a real memory.

The processes that produce revenue are reasoning habits, not workflows. Borrowing Wolfram's computational reducibility to find the pockets inside them — and to let the record assemble itself.

A business world model, the org chart as a routing protocol, the end of manual CRM entry, and the end of hardcoding. Read together, they describe one architecture — and one uncomfortable org change.

For decades we built corporate software focused on one thing: keeping a record. The next era is about reasoning over context.

Most modern SaaS platforms are graveyards of disconnected data. A context graph reconnects it.

Notes on the technical side and business application of the context graph concept, and where the open research questions are.