Semantica
The traceable semantic layer under your AI agents
About
Semantica is an MIT-licensed semantic and context layer that sits underneath your LLM, vector store and agent framework. It ingests fragmented multi-source data, extracts entities, relations and events, builds a queryable Context Graph and knowledge graph, and makes the meaning of that data explicit through ontologies and controlled vocabularies (OWL, SHACL, SKOS) rather than leaving it as an embedding. After pip install semantica you drive it from Python: every agent decision becomes a first-class graph object you can trace back through its causal chain, search by precedent, analyse for downstream impact, and gate against policy rules. What separates it from vector search plus RAG is recall and explainability — graph traversal returns a structured path, forward chaining, Rete, Datalog and SPARQL reasoning are fully deterministic, no LLM call is needed to build the graph or its provenance, and provenance itself is recorded in W3C PROV-O and exportable to JSON, CSV or RDF. Graph storage is polyglot: RDF triple stores (embedded Oxigraph, Blazegraph, Apache Jena, Eclipse RDF4J) and labeled property graphs (Neo4j, FalkorDB, Apache AGE, AWS Neptune) are interchangeable without touching your code. It also ships an MCP server, a CLI, a REST API, native Agno / CrewAI / LangChain integrations, and a browser-based Knowledge Explorer for walking the graph, ontologies and timelines interactively.
Key Features
- ✓Ingest multi-source data, extract entities, relations and events
- ✓Record every AI decision as a queryable, traceable graph node
- ✓Govern ontologies and detect conflicts via OWL, SHACL, SKOS
- ✓Reason deterministically with forward chaining, Rete, Datalog, SPARQL
- ✓Track provenance in W3C PROV-O and export audit trails
- ✓Ship an MCP server, CLI, REST API and LangChain / CrewAI integrations
Who it's for
- AI platform engineers who need structured context for agents, not just a vector index
- Compliance, risk and audit teams who must answer "why did the AI decide that?"
- Not for users wanting something turnkey — this is infrastructure you self-host and drive from Python
Pros & cons
Pros
- Graph traversal returns a structured "why" path instead of a bare similarity score
- Graph building, reasoning and provenance are deterministic and LLM-free, so results reproduce
- Graph backends are swappable (RDF stores, Neo4j, Neptune) with no code changes and no vendor lock-in
- It complements rather than replaces — keep your existing LLM, vector store and agent framework
Limitations
- Still 0.x (0.6.8 on PyPI) with APIs in flux — 0.7.0 moved heavy deps into optional extras, changing your install command
- Production means your own Docker / K8s plus a persistent graph store; a local pip install is trial-only, and managed deployment with SLA support sits in an Enterprise plan with no public pricing
- Getting real value from ontology governance means learning OWL / SHACL / SKOS modelling — a far steeper curve than adding a vector store
- It explains only what is outside the model — context in, decision out, execution trail; the LLM's own reasoning stays opaque, as the project states