Semantica

Semantica is a semantic and context layer that sits underneath an LLM, vector store, and agent framework. It turns fragmented source data into a queryable context graph and knowledge graph governed by ontologies and controlled vocabularies (OWL, SHACL, SKOS), so relationships and meaning are explicit rather than implied by embeddings. Graph construction, reasoning, and provenance are deterministic and need no LLM.

The scope of what it explains is stated narrowly: Semantica does not reconstruct what happens inside a model, and chain-of-thought stays opaque. What it records is everything outside the model, as first-class objects: the context fed in, the decision produced, its provenance, the relationships involved, the policies applied, and the execution trail. Decisions become queryable nodes, so you can trace a causal chain, find precedents among similar past decisions, analyse downstream impact, and check a decision against policy rules.

The pipeline is a sequence of independently importable modules: ingest, parse, normalize, split, extract, conflict detection, deduplication, knowledge-graph construction, then an intelligence layer of ontology, reasoning, provenance, and decisions, and finally storage, export, and access. Ingestion covers files, web pages with robots.txt compliance, SQL databases, Parquet, Git, email, cloud sources, streams such as Kafka and Kinesis, and enterprise platforms including Databricks Unity Catalog, Snowflake, and SAP OData, so warehouse tables become graph nodes with lineage instead of a CSV export.

Storage is polyglot and swappable without code changes: RDF triple stores (embedded Oxigraph, Blazegraph, Apache Jena, Eclipse RDF4J over SPARQL), labeled property graphs (Neo4j, FalkorDB, Apache AGE, AWS Neptune over Cypher), and vector stores. Reasoning offers forward chaining, a Rete network, Datalog, and SPARQL with explainable paths, and graph analytics adds centrality, community detection, link prediction, and shortest-path queries. Provenance follows W3C PROV-O and exports to JSON, CSV, or RDF, with bi-temporal facts and point-in-time snapshots of the graph.

Access paths include native support for the Agno, CrewAI, and LangChain agent frameworks, an MCP server, a REST API, a command-line interface, and plugin bundles for several coding agents and editors. Model calls that do need an LLM go through LiteLLM, so the usual hosted and local providers are available.

Features

• Context graphs: typed, weighted graph nodes and edges with traversal, snapshots, and an agent-memory API
• Decision intelligence: decisions as first-class objects with causal chains, precedent search, and impact analysis
• Deterministic reasoning: forward chaining, Rete, Datalog, and SPARQL with explainable inference paths
• Ontology governance: SHACL constraints, OWL generation, SKOS vocabularies, and conflict detection
• Provenance: W3C PROV-O lineage on every fact, exportable to JSON, CSV, or RDF
• Knowledge pipeline: multi-source ingestion, entity-aware chunking, NER, relation and event extraction, deduplication
• Enterprise connectors: Databricks Unity Catalog, Snowflake, and SAP OData with lineage introspection
• Polyglot storage: RDF triple stores, labeled property graphs, and vector stores, swappable without code changes
• Graph analytics: centrality, community detection, link prediction, and shortest-path queries
• Bi-temporal facts: point-in-time snapshots of the graph as it stood on any past date
• Framework integrations: native Agno, CrewAI, and LangChain support plus an MCP server and REST API
• Editor plugins: bundles for several coding agents and editors, with a browser workbench for visualization