I'm excited to announce a new release of OntoCast — an open-source framework for extracting semantic triples and building knowledge graphs (KG) from unstructured documents (PDF, JSON, Markdown, and more).

Before extracting facts, OntoCast automatically selects or creates a relevant ontology and iteratively refines it, leading to much more accurate and context-aware fact extraction. This is especially valuable for cross-domain or complex documents where a static ontology falls short.

- Agentic workflow: Uses LLMs (OpenAI/Ollama) to drive the extraction and ontology refinement process.

- MCP-compatible API server: Easy to integrate into your stack.

- Flexible storage: Works with Jena Fuseki and Neo4j for knowledge graph storage.

- Open source: Apache licensed.

Uses cases include extracting structured knowledge from scientific papers, financial reports, or clinical trial documents — even when they span multiple domains.

Repo: https://github.com/growgraph/ontocast Docs: https://growgraph.github.io/ontocast

Would love feedback, questions, or suggestions!