Knowledge graphs (GraphRAG)

Documents are flat; the things in them are connected: this customer uses that feature, which this ticket is about, which that decision changed. Graph engineering stores the connections explicitly, as a knowledge graph, so the assistant can follow relationships and answer questions that need several hops. GraphRAG and graph memory are its common forms. Since mid 2026 the name graph engineering is also used for arranging agents and tasks as a graph; this page covers the knowledge side.

Standing: Established since 2024-07
Stars, week: +1,182 across 6 repositories
Stars, month: +7,660, through 2026-11-08
Listings: 6 chosen by hand

The standing

Where it stands

Among the 14 concepts. Bars: GitHub stars gained this week.

Established · 7

Rising · 7

How to use it today

In ChatGPT or Claude, as a person doing the work, not as a developer.

1Notice when your questions are relational ("which customers who asked for X also churned"): those are graph questions, and flat retrieval answers them badly.
2Start without a database: ask the assistant to extract entities and relationships from a set of documents into a table, and keep that table as your graph.
3For a product that needs it, use a graph memory or GraphRAG tool from the catalog and point it at a bounded document set first.
4Name the entity types that matter (customer, feature, decision, person) before extracting; everything else is noise.
5Keep the graph small and current; a stale graph answers with confidence about relationships that no longer exist.

What changes day to day

Multi-hop questions become answerable in one ask. Your work shifts to naming the entities and relationships your organization actually reasons about.

Where it goes wrong

Most teams do not need a graph; they need cleaner documents.

Automatic extraction produces plausible but wrong edges, which then get cited.

Listings that use it well

Chosen by hand after reading them, each with the reason. Install from the listing's own page.

Listing	Kind	Why it is here	Stars
ai-agent-book	Agent skills	knowledge-org covers RAPTOR and GraphRAG index choices and graph structured retrieval.	57k
Codebase Memory	MCP server	A code knowledge graph for 162 languages with sub millisecond queries.	48k
code-review-graph	Agent skills	Builds a persistent code intelligence graph and reviews PRs with blast radius from it.	33k
CodeGraphContext	MCP server	Indexes local code into a graph database and serves it over MCP.	4.4k
mcp-knowledge-graph	MCP server	Persistent memory for Claude held as a local knowledge graph.	888
graph-engineering	Agent skills	Teaches ontology design, entity and relation extraction, fusion and GraphRAG serving, plus task graphs for orchestration.	567

Elsewhere

Repositories that embody the idea and are not in the catalog.

microsoft/graphrag ↗The reference graph-based RAG pipeline: entity extraction, community clustering, summarization.
HKUDS/LightRAG ↗Lighter graph plus vector retrieval (EMNLP 2025).
getzep/graphiti ↗Temporal knowledge graph engine for agent memory, the core of Zep.

Standing

Our judgment, with the reason and the sources it rests on.

Established since 2024-07. GraphRAG and graph memory have had working implementations since 2024; building and maintaining the graph is the cost. Microsoft's GraphRAG paper (April 2024) and open-source release (July 2024) made graph-based retrieval a known technique, and Zep's temporal graph (January 2025) carried it into agent memory. Settled practice: the methods are known, the tools are mature, and the question is how well you use it.