shacl2cypher
Validate a property graph with shapes, not hand-written queries.
Compiles W3C SHACL shapes into named, read-only Cypher that checks Neo4j, LadybugDB and FalkorDB and points at every broken rule.
Memory for agents and humans.
My open-source work, in two lines: memory systems — the shared spaces where agents and people keep context and exchange knowledge — and modelling tools for drawing a domain and shipping it as text.
Memory & agents
Open-source memory for agents and the teams around them: a shared graph that people and agents write together, a database that never forgets what it learned or when, and memory that stays on the device. The research behind it is on the Papers page.
TypedGraph
A shared space for agents and humans.
Turns Markdown notes and code into a typed, validated knowledge graph that people and AI agents read and write together.
OxigraphDB
A graph database small enough to live where the agent lives.
An embeddable RDF / SPARQL graph database, built as the knowledge and memory layer for agentic AI running at the edge.
Tiramemsu
Layered, never-forget memory for agents.
An embedded graph database on SQLite in which every fact has an id, so facts carry provenance and belief — and every change is kept with when it was made and when it was true.
Draw the model. Ship the text.
Three VS Code extensions, one conviction. You model visually because a diagram is how a person understands structure — and then you commit plain, schema-validated JSON, because text is what diffs, merges, reviews and generates. The picture is a projection you can throw away and redraw. The model is the asset.
Factum ORM
Conceptual schemas in ORM 2 — drawn, verbalized, and checked.
Object-Role Modeling describes a domain as elementary facts — *Person works for Company* — instead of tables or classes. Because those facts are attribute-free, every constraint is explicit and the whole model reads back as plain English sentences a domain expert can confirm or reject. Draw the schema, read the FORML verbalization, validate it, then map it to a relational or a property graph schema. In the spirit of NORMA, but native to VS Code.
LPG Modeler
Design a property graph once. Generate every schema from it.
Author a Labeled Property Graph schema as text, view it as an ERD-like diagram, and generate the artifacts from that single model: LadybugDB DDL, Neo4j constraints, SHACL shapes, and an OWL ontology. One source of truth instead of four drifting ones. VS Code extension and CLI.
Causal Canvas
Causal models that tell you when you are wrong.
A visual editor for causal models whose real output is a plain, schema-validated JSON file — not a picture. It also knows what a causal model means: it will tell you when you are adjusting for a collider, when an instrument violates the exclusion restriction, and when your exposure has no path to your outcome at all.
Graph tooling
A way to prove a property graph still matches its model, built in the open.
Research & reference
Libraries, reference implementations and research — smaller than the tools above, kept public because the results are more useful in the open.
- agentic-memory 11★
Multi-layered memory architectures for agents — episodic, semantic, procedural.
- ladybug-rag 9★
Reference implementation of Hybrid Graph RAG with LadybugDB, in Python.
- ladybug-rag-rs 5★
Four retrieval modes in one query — vector search, graph traversal, PageRank and community detection. +109% on multi-hop questions against vector-only RAG.
There is a good deal more at github.com/Volland — ontology editors, graph engines, and a quantity of half-finished ideas.