Cograph
DemoGitHub

Quickstart

From zero to querying your data in 5 minutes.

1. Start the graph database

docker compose up -d

This starts Apache Fuseki on port 3030. Data persists in a Docker volume.

2. Install Cograph

python -m venv .venv
source .venv/bin/activate
pip install -e .

3. Configure

cp .env.example .env
# Edit .env and add your OpenRouter API key:
# OPENROUTER_API_KEY=sk-or-...

4. Start the API server

source .env
uvicorn omnix.api.app:create_app --factory --port 8000

5. Ingest a CSV

omnix ingest your-data.csv --kg my-dataset

One LLM call infers the schema from your column headers and sample rows. All rows are then mapped deterministically. No LLM per row.

6. Ask questions

omnix ask "How many records are there?" --kg my-dataset
omnix ask "Which category has the highest average price?" --kg my-dataset

What happens under the hood

Ingestion: The CSV columns are analyzed by an LLM to determine which are attributes (numbers, dates, booleans) and which are relationships to other entities (cities, companies, categories). A typed ontology is created automatically. Each row becomes an entity with typed triples stored in the graph database.

Querying: Your natural language question is translated to SPARQL using the ontology as context. The query runs against the graph database and results are formatted as a human-readable answer.

Next steps