Skip to content

Ontology Studio

Ontology Studio is where you define the schema for a knowledge graph — the node types, relationship types, and properties that your data will be structured into. A defined ontology constrains LLM extraction so you get typed, validated entities instead of generic "entity soup".

Find it under Build → Ontology (/build), or open a graph's ontology directly.

Creating an ontology

  1. Click Create Ontology.
  2. Give it a name.
  3. Select the Knowledge Graph it applies to (only graphs in ready state appear in the dropdown).
  4. Click Create — you start with an empty schema.

Each ontology card shows its KG, version, and counts: 3 nodes · 5 rels.

Defining node types

The left panel lists Node Types. To add one:

  • Type a name in the "New type…" box and press Enter (or click +).
  • Click a node type to select it and edit its properties.
  • Hover and click × to remove a type — this also removes any relationships that reference it.

Tier limits

The number of node types per ontology depends on your plan. When you hit the limit, the add button is disabled and a counter shows N/max — upgrade for more. See Pricing.

Defining relationships

Relationships connect a source node type to a target node type. The center canvas renders your schema as a live graph — nodes arranged in a ring with directed arrows for each relationship. Click a node to highlight its connections.

Import & export

Ontology Studio interoperates with standard ontology formats:

  • Import — load an existing schema from OWL or JSON-LD. Useful for reusing a domain ontology (e.g. from Protégé).
  • Export OWL — download your schema as OWL for use in other RDF tools.

Tier gating

Import and export are gated by plan (allow_ontology_import / allow_ontology_export). If disabled, the buttons show an "Upgrade to unlock" tooltip.

Why it matters

A guided ontology is what separates Graffold from generic RAG:

  • Extraction produces entities that match your schema, not arbitrary blobs.
  • Relationships are typed, so multi-hop traversal is meaningful.
  • Exported OWL/JSON-LD keeps you portable — no lock-in.

See Architecture → Semantic Dynamic Context Graph for how the ontology feeds the extraction and query pipeline.