Graph Engineering

The Loop Memory Model

Give agentic workflows durable memory, traceability, and shared knowledge with graphs.

Agent loops are good at producing work, but the reasoning behind that work disappears the moment the context window closes. The next run starts from a transcript, or from nothing, and the decision that mattered last week is rediscovered — or contradicted — this week.

The book moves decisions, evidence and execution history out of transcripts and into an explicit graph. Markdown stays the source of truth; the graph is a query index you can rebuild, not a second copy that drifts. From there it walks through parsing a repository into the graph, resolving entities, retrieving small neighbourhoods within a token budget, checking claims against graph paths, coordinating parallel workers, and tracing lineage from an objective to the artifact that met it.

It ends on judgement: how to measure whether the graph is any good, how it fails, what it costs — and when an ADR log or a relational table is the better answer.

Buy on Amazon On Leanpub: Graph Engineering: The Loop That Remembers