No Monkey
Business
← All notes
Infrastructure2 min read321 words

Ontology and knowledge graph tools for AI agents: a roundup

Ontologies and knowledge graphs are a hot topic again. Here are the tools worth saving.

On this page

It’s been an eventful few months in ontology and knowledge graph technology. Apparently it’s officially a hot topic again. Save these for later.

Under the hype, almost every tool below is doing the same job: turning raw material into a typed graph an agent can actually traverse, instead of a pile of vectors it can only similarity-search.

If that sounds abstract, here is the whole idea with monkeys and bananas. Watch it loop: a pile of bananas is just raw data; an ontology writes down the types and relations; the graph instantiates them; and a monkey’s question gets answered by traversal, not guesswork.

NMB · FIG.01 · ANIMATED ONTOLOGY, WITH MONKEYS & BANANAS A pile is not knowledge. pile —model→ schema —instantiate→ graph —query→ answer 1THE PILE raw data · no structure 2THE ONTOLOGY Banana —isA→ Fruit Banana —hasColor→ Yellow Monkey —eats→ Banana typed classes + relations the schema agents share 3THE GRAPH b1 b2 b3 Fruit Yellow instances you can traverse 4THE ANSWER which are ripe? ripe = b2 & b3 traverse, don’t guess 1PILE 2SCHEMA 3GRAPH 4ANSWER
The whole idea, animated: pile → schema → graph → answer. The monkey traverses; it doesn't guess.
NMB · FIG.02 ONTOLOGY & KG TOOLING · THE SHARED PATTERN Give agents a graph, not a pile. source —extract→ ontology —instantiate→ graph —query→ agent vector-only RAG skips the structure 01 SOURCE docs · metadata · code the raw material 02 ONTOLOGY typed entities + relations the schema agents share 03 GRAPH queryable · provenance memory you can traverse 04 AGENT retrieve · reason · code acts on structure WHAT TO LOOK FOR local / offline extraction provenance on every fact a typed, versioned schema Every tool in this roundup sits at one of these four stages.
The shared pattern. Structure in the middle is what separates a graph agent from vector-only RAG.

Ontology layers for AI agents

Ontology engineering and specs

Graph memory for agents

Codebase graphs for coding agents

The whole shelf at a glance

NMB · FIG.03 ONTOLOGY & KG TOOLING · THE LANDSCAPE Four shelves worth saving. ten tools · four jobs · one idea 01 ONTOLOGY LAYERS a semantic layer over enterprise data NVIDIA Auto Ontology · AWS Context Ontology OpenBKN · grounding for text-to-SQL + agents 02 ENGINEERING & SPECS validate, reason, roll back; define the graph Open Ontologies · Neo4j Ontology Graph Spec YAML/JSON format + change control 03 GRAPH MEMORY embedded / offline memory with provenance Relate · cognee · Semantica cognee extracts with GLiNER2, no LLM needed 04 CODEBASE GRAPHS AST → queryable graph for coding agents Ontology Atlas · Graphify (YC S26) local parsing, no vector store in the loop Bookmark the shelf, not just the tool.
Ten tools, four jobs. Bookmark the shelf, not just the tool.

Originally shared on LinkedIn. Adapted with the author’s permission.