Semantica

Semantica is an open-source Python library (MIT licensed) that sits underneath LLMs, vector stores, and agent frameworks as a deterministic infrastructure layer for context graphs, decision intelligence, and full provenance tracking. It is installable via pip install semantica and is currently at v0.6.0, published July 2026. The project describes itself as "the glass-box alternative to black-box intelligence" and is positioned for high-stakes, regulated domains where AI outputs must be explainable and auditable.

What It Is

Semantica is a graph-native AI infrastructure library that builds a structured, queryable context graph from enterprise data sources and records every AI decision as a first-class, auditable object. Unlike vector stores that retrieve similar text without understanding relationships, Semantica constructs a live knowledge graph of entities, relationships, and decisions with W3C PROV-O compliant lineage on every fact. It is fully self-hostable with zero vendor lock-in, supporting swappable backends for graph databases (Neo4j, FalkorDB, Apache AGE, AWS Neptune), RDF triple stores (Blazegraph, Apache Jena, Eclipse RDF4J), and vector stores (FAISS, Pinecone, Weaviate, Qdrant, Milvus, PgVector).

Core Architecture and Modules

Semantica ships 13+ independently importable production modules spanning the full pipeline from ingestion to export:

• Context Graphs — Typed nodes and edges with temporal validity windows, SPARQL/Cypher queries, and point-in-time snapshots
• Decision Intelligence — record_decision(), trace_decision_chain(), find_similar_decisions(), analyze_decision_impact(), and check_decision_rules() turn every AI choice into a permanent, queryable graph node
• Reasoning Engines — Forward chaining, Rete network, Datalog, abductive reasoning, and SPARQL with fully explainable inference paths
• Semantic Extraction — Named entity recognition, relation extraction, event detection, and triplet generation from raw text
• Provenance & Auditability — W3C PROV-O lineage on every fact, exportable to JSON, CSV, RDF, and Parquet for regulator submission
• Ontology Management — Auto-generate OWL ontologies, validate with SHACL/HermiT/Pellet, manage SKOS vocabularies
• Temporal Intelligence — Allen Interval Algebra (13 relations), bi-temporal provenance, point-in-time graph snapshots, and TemporalNormalizer
• Pipeline DSL — Declarative, parallel stage chaining with validation gates and configurable retry policies

Integration Ecosystem

Semantica connects to virtually every major AI tool category through native plugin bundles, an MCP server, and a REST API:

• Native plugin bundles for Claude Code, Cursor, and Codex CLI
• MCP server + plugin for Windsurf, Cline, Continue, VS Code, and OpenClaw
• REST API for GitHub Copilot, Roo Code, Goose, Aider, Amazon Q, and Zed
• Agentic frameworks — First-class Agno integration; LangChain, LangGraph, CrewAI, LlamaIndex, AutoGen, OpenAI Agents, and Google ADK supported via REST/MCP today, with dedicated native SDKs planned
• Enterprise data platforms — Native DatabricksIngestor (Unity Catalog + Delta Lake) and SnowflakeIngestor connectors
• LLM providers — All major providers via semantica.llms and LiteLLM: OpenAI, Anthropic, Gemini, Mistral, Llama, Groq, Cohere, Azure OpenAI, AWS Bedrock, Ollama, DeepSeek, and more

Target Audience and Regulated Domains

The README explicitly targets AI/ML platform teams shipping consequential agents, data platform teams on Databricks or Snowflake, compliance and audit teams, and regulated enterprises in finance, healthcare, legal, government, and defense. The homepage lists vendor-published case studies including a marine natural capital finance deployment (Nereus) quantifying $1.62B in natural capital across Indo-Pacific sites, and an energy regulation deployment (Ausgrid) indexing 10 years of AER regulatory history with full provenance. The project states compliance alignment with GDPR, EU AI Act, HIPAA, and W3C PROV-O standards.

Update: v0.6.0

Released July 21, 2026, v0.6.0 adds four notable capabilities: named-graph support for JenaStore completing cross-backend named-graph parity across Blazegraph, RDF4J, and Jena; parameterized SPARQL CONSTRUCT query templates extended to RDF4J and Jena with pipeline integration; a new DatabricksIngestor for Unity Catalog + Delta Lake ingestion with PAT/OAuth M2M auth; and a SQLiteVecStore disk-backed local vector store on sqlite-vec's vec0 virtual tables. The repository reports 3,100+ GitHub stars and 370+ forks as of the data snapshot.