OntoCast
Open source · Apache-2.0Agentic LLM knowledge graph builder for semantic triple extraction from documents
Documents → auditable knowledge graph, guided by your ontology.
An agentic framework for semantic triple extraction: LLM-powered reading, ontology-guided consistency, RDF you can audit — every fact grounded in its source.
$ pip install ontocast
Why OntoCast
Extraction you can defend, not just demo.
Vector search retrieves text that looks similar. OntoCast builds an explicit model — typed entities and relations your ontology defines, with provenance attached.
Ontology-guided, not free-form
Extraction is steered by your vocabulary, so entities and relations come out consistent with the model you already govern — not whatever the LLM improvised that day.
Every fact has a source
Extracted triples stay grounded in the passage they came from. When someone asks where a statement in the graph comes from, there is an answer.
Entities resolved across chunks
The same supplier mentioned under three different names across forty pages becomes one entity, not three — disambiguation is part of the pipeline.
Standard RDF out
Output is RDF/Turtle against standard vocabularies, ready for any triple store — nothing about the result needs OntoCast to remain readable.
Documents in, many formats
PDF, Word, PowerPoint, Excel, HTML, Markdown, images, JSON and plain text flow through semantic chunking sized for LLM processing.
Built for pipelines
API endpoints slot into RAG systems and agentic workflows; the managed OntoCast API Platform runs the same engine as a service.
Use cases
Where it fits
GraphRAG
Ground retrieval-augmented generation in structured facts instead of similarity-ranked text chunks.
Ontology population
Keep a governed vocabulary filled from the documents your organization already produces.
Domain knowledge graphs
Finance, research, compliance — build graphs where auditability of each fact matters.
One pipeline
Documents are half the story
OntoCast reads what's unstructured. GraFlo composes what's already structured — SQL, CSV, APIs — into the same governed graph. The managed OntoCast API Platform runs the pipeline for you.
Resources