Build resilient systems that
scale across agents
Graph Engineering transforms how distributed systems handle complexity. Instead of fragile file-based states, it uses resilient graph structures where multiple workers can coordinate safely.
See it in action
Watch Graph Engineering initialize a dual-agent memory system in real time — from project scaffolding to a fully built graph structure.
Two graphs, one system
Graph Engineering separates what was attempted from what turned out to be true. Every agent reads the same verified facts — none of them replay each other's trial and error.
Work History
Every attempt an agent makes is recorded in the work-history graph. Previous attempts are marked superseded — agents never act on stale work.
Verified Facts
The fact graph holds only what has been verified true. Every edge carries a receipt — agents can trust the data without re-verifying it themselves.
Work history records attempts. Facts record what is true. Agents read facts; they do not replay history.
The lifecycle of a fact
A fact is not created all at once. It passes through three stages — each one adding confidence, context, and a verifiable trail. This is how raw input becomes structured, trustworthy knowledge.
Extraction
Raw input is parsed into structured facts. Unstructured prose becomes typed, queryable triples that downstream systems can reason about.
Resolution
New facts merge with existing ones. Contradictions are flagged rather than silently overwritten, preserving the full decision trail.
Provenance
Every fact carries a receipt — who created it, when, and under what context. Auditors and agents can trace any claim back to its source.
Extraction
Raw input is parsed into structured facts. Unstructured prose becomes typed, queryable triples that downstream systems can reason about.
Resolution
New facts merge with existing ones. Contradictions are flagged rather than silently overwritten, preserving the full decision trail.
Provenance
Every fact carries a receipt — who created it, when, and under what context. Auditors and agents can trace any claim back to its source.
Seventeen steps, seven parts
Master Graph Engineering through a carefully structured curriculum. Steps 1–13 build the foundation with queryable fact graphs and worker coordination. Steps 14–17 explore advanced patterns for scaling and optimization across distributed systems.
Core Patterns
Twenty-three battle-tested patterns covering extraction, resolution, provenance, governance, and storage. Each pattern addresses specific challenges in building reliable graph systems that scale across multiple agents.
Core Patterns
Essential solutions that form the foundation of Graph Engineering.
Explore all 23 patterns on the patterns page.
Get Started
Follow four simple steps to begin your Graph Engineering journey. From initial setup to running your first starter kit, you'll have a working example in minutes.
Read Start here
Two or three questions about what you have already built, and it points you at the page to open first.
Pick a track
G1 through G4, beginner to expert. Each finishes where the next one assumes you are.
Install the core package
Get the foundation package that powers all Graph Engineering projects.
$ npm install @graph-engineering/coreClone a starter kit
One command, no API keys, no build. Run this in your terminal:
$ npx @graph-engineering-kits/graph-kit document-to-facts