Semantica Is an Open-Source Audit Trail for AI Decisions

Every decision your AI agent makes is currently disappearing into a chat log. When a compliance officer asks "why did it do that?" months later, you have no answer.

Semantica is an open-source layer that records each decision as a queryable graph node — with reasoning, outcome, and confidence included — so the full story is always retrievable.

What Happened

Semantica builds a Context Graph where entities, relationships, and decisions are all nodes. Each decision stores its reasoning, outcome, and confidence score, and links to other decisions through explicit causal relations like CAUSED, INFLUENCED, and PRECEDENT_FOR. That means you can trace the full causal chain of any decision, search past decisions as references, see what later got affected, and compare everything against your policies.

Every record also carries W3C PROV-O provenance data, showing where the information came from and how it was produced. You can export the whole chain to hand to a regulator or an auditor.

You don't need to change your stack. Semantica sits underneath your existing LLM, vector store, and agent framework. Storage is flexible: RDF triple stores like Oxigraph, Blazegraph, Apache Jena, and Eclipse RDF4J, or property graphs like Neo4j, FalkorDB, and AWS Neptune — all swappable without touching your code. It also connects directly to Databricks and Snowflake, so data already in your lakehouse or warehouse becomes graph nodes with provenance intact.

The project is MIT licensed, has more than 5.2k stars, and ships with a CLI, REST API, and MCP server. There are ready-made plugins for Claude Code, Cursor, Codex, Windsurf, and others.

Why It Matters

Instead of an AI agent that decides and forgets, you get a clear record that can answer months later: what it decided, why, what it affected, and where the information came from. That's exactly the infrastructure you'll need as agents start making real decisions inside companies.