Quick Start

This guide walks you through generating synthetic RDF data for a minimal ontology.

You need a JDK 25 build and a working LLM backend. See Installation first.

1. Prepare your ontology

Create input.ttl — a minimal OWL ontology in Turtle format:

@prefix owl:  <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix ex:   <http://example.org/> .

ex:Person a owl:Class ;
    rdfs:comment "A human being." .

ex:Organization a owl:Class ;
    rdfs:comment "A legal organisation." .

ex:worksFor a owl:ObjectProperty ;
    rdfs:domain ex:Person ;
    rdfs:range  ex:Organization .

2. Create the configuration

Create config.yml next to it:

input-path: "input.ttl"
output-path: "output.ttl"
user-input: "Generate 5 example persons each working for one of 2 organisations."

client:
  provider: "copilot"
  worker:
    model: "gpt-5-mini"
    timeout-ms: 120000
    batch-size: 2
    pool-count: 1
  supervisor:
    model: "gpt-5.4"
    timeout-ms: 120000
  reviewer:
    model: "claude-4.6-sonnet"
    timeout-ms: 60000

provider: "copilot" is the default and drives the GitHub Copilot CLI. Switch it to ollama or openai-compatible if you would rather run against a local or self-hosted model — see the Configuration Reference.

WARNING

OWL-SDA deletes the file at output-path before every run. Point it at a fresh file, not at something you want to keep. A real run also spends real tokens on whichever backend you configured, so start small.

3. Build and run

Build the executable jar (requires JDK 25):

mvn -DskipTests clean package

Then run it against your configuration. --config is required — running the jar without it prints a usage error rather than picking a default:

java -jar target/owlsda.jar --config config.yml

Two more things the CLI gives you:

java -jar target/owlsda.jar --help      # or -h: full option list
java -jar target/owlsda.jar --version   # or -V: prints 1.0-SNAPSHOT

--config accepts a plain path, or an explicit prefix:

Value	Resolution
config.yml	Classpath first, then the filesystem.
file:config.yml	Filesystem only.
classpath:config.yml	Classpath only.

See the CLI Reference for every option.

4. Inspect the output

The triple store is flushed to output.ttl at each delegation round, before the supervisor inspects it, so you can watch intermediate progress:

cat output.ttl

To watch the run in a browser instead, enable benchmarking in your config and start the dashboard:

benchmark:
  enabled: true

java -jar target/owlsda.jar --config config.yml --web-ui

It serves on http://localhost:8080 (change it with --web-ui-port) and stays up after the run finishes until you press Ctrl+C. See the Web UI guide.

Next Steps

• Run full, ready-to-use Examples (including benchmark visuals).
• Read the full Configuration Reference to tune model selection, timeouts, and pool sizes.
• Enable Benchmarking to track token usage and violation counts per round.
• Add External Ontologies to resolve owl:imports automatically.