semantica: Auditable Decision Path Tracing Without LLMs

Most AI agents store only embeddings without preserving meaning or context, making it impossible to explain or audit their decision-making processes. Semantica is a deterministic infrastructure layer positioned beneath LLMs, vector stores, and agent frameworks, requiring no LLMs for graph construction, reasoning, or provenance generation. It provides system-level explainability by clearly tracing the origins of input data and output decisions, rather than the reasoning process inside the model.

It collects enterprise data to extract entities and relationships, detects conflicting facts, and constructs knowledge graphs. With native connectors for Databricks and Snowflake, it converts tables from existing data warehouses into graph nodes with provenance without requiring separate exports. It supports both RDF and Labeled Property Graphs, allowing backend swaps among various options such as Oxigraph, Neo4j, and Apache AGE without code changes.

Unlike traditional RAG approaches that retrieve based solely on embedding similarity and leave no decision history, Semantica treats every decision as a first-class object. It assigns provenance according to the W3C PROV-O standard and performs reasoning through explainable paths using Forward chaining, Rete network, Datalog, and SPARQL. It enforces compliance rules via SHACL constraints and OWL generation, and allows exporting audit trails in JSON, CSV, and RDF formats.

It is suitable for regulated industries such as finance, healthcare, legal, and defense, where deploying black-box AI is not permitted. Self-hosting ensures data does not leave for third-party SaaS, and the stack can be replaced without vendor lock-in. It offers an MCP server, REST API, and CLI, and integrates natively with Agno and CrewAI for immediate application in existing agent workflows.