OntoCast

Agentic ontology-assisted framework for semantic triple extraction

---

Overview

OntoCast extracts semantic triples from documents using an agentic, ontology-driven pipeline. It co-evolves ontologies and facts graphs with parallel per-chunk processing, RDF 1.2 provenance, and optional vector-backed ontology retrieval.

---

Key Features

• Parallel map/reduce pipeline — concurrent per-unit ontology and facts loops
• Robust entity disambiguation — embedding + symbolic alignment across chunks
• RDF 1.2 provenance — quoted triples, provenance artifacts, optional strip_provenance
• GraphUpdate operations — token-efficient SPARQL insert/delete instead of full graph regeneration
• JSON-LD wire format — optional LLM_GRAPH_FORMAT=jsonld for LLM payloads
• Ontology context modes — catalog selection, vector retrieval, or fixed ontology
• Triple store integration — Fuseki, Neo4j (n10s), or filesystem fallback
• Tenancy — partition datasets/collections by tenant and project
• REST API — document processing, ontology catalog management, graph matching
• Automatic LLM caching — built-in response caching

---

Documentation

• Quick Start
• Workflow
• Core Concepts
• Configuration
• API Endpoints
• Tenancy
• Ontology Context
• Triple Stores
• LLM Caching
• API Reference

---

Installation

uv add ontocast
# or
pip install ontocast

Optional PDF/DOCX conversion: pip install "ontocast[doc-processing]"

---

Quick Start

cp .env.example .env
# Edit LLM_API_KEY and paths

ontocast --env-path .env

curl -X POST http://localhost:8999/process -F "file=@document.pdf"

See Quick Start Guide for full configuration.

---

REST API (Summary)

Method	Path	Purpose
GET	/health	Health check
GET	/info	Service metadata
POST	/process	Full document pipeline
POST	/process_unit	Single content unit
POST	/flush	Clear triple store data
POST	/ontologies	Upload catalog ontology
PUT/DELETE	/ontologies/{iri}	Replace or delete ontology
POST	/match/entities	Global entity alignment
POST	/match/derive-matches	Pairwise entity matching
POST	/match/evaluate	Triple/entity metrics

Details: API Endpoints.

---

Workflow

Document-level pipeline (regenerated via uv run plot-graph):

Landscape variant: graph.lr.png. Per-unit render/critic loops are documented in Workflow.

1. Convert → chunk document
2. Parallel ontology render per unit → normalize → optional consolidate → validate
3. Parallel facts render per unit → merge with disambiguation
4. Serialize to triple store; return Turtle in API response

---

Project Structure

ontocast/
├── agent/           # Render, critic, normalize, serialize agents
├── api/             # FastAPI routers (ontologies, schemas, tenancy)
├── cli/             # Server and utility CLIs
├── onto/            # Ontology, RDFGraph, state models
├── prompt/          # LLM prompt templates
├── stategraph/      # LangGraph workflow
├── tool/            # Triple stores, chunking, vector store, aggregation
├── config.py        # Pydantic settings
└── toolbox.py       # Tool dependency container

---

Contributing

See Contributing and CHANGELOG.