Technology · Knowledge Fabric

Why geometry?

Underneath the GeometriX agentic products is a shared substrate: knowledge represented geometrically, retrieved deterministically, verified mathematically — and connected to whatever stores you already run.

GMS — the Geometric Memory System Stable

A knowledge graph embedded on a high-dimensional sphere: entities become points, relationships become rotations. Retrieval is deterministic, contradictions create measurable tension, and retrieved facts carry integrity verification. Patent pending.

Deterministic retrieval

Run the same query against the same store twice and you get the same evidence both times. That repeatability is what lets retrieval survive an audit.

Contradiction detection

Conflicting knowledge produces geometric tension the system can measure and surface.

Verified facts

Integrity checks on every retrieved fact; LLM outputs that contradict the store are rejected.

DocGMS — documents in, expert system out Available

Ingest documents into a geometric expert store, query it deterministically, verify LLM responses against it, and let verified outputs expand the store over time. Served to any agent over MCP by the built-in servers.

from knowlytix.knowledge import GMSExpertStore, QueryEngine, ingest_document

KAL — the Knowledge Adapter Layer Beta

A backend-agnostic mesh for knowledge graphs: one typed protocol, a federation router, and adapters for Neo4j, PostgreSQL + pgvector, SPARQL endpoints, vector stores, files and external MCP sources — presented as a single logical graph with verification metadata on everything that crosses the boundary.

One protocol

Typed nodes, triples, literals, queries and results — with confidence, status and evaluator identity attached.

Federation

Multiple backends behind one router, with explicit conflict strategies.

Multi-tenant

Tenant indexing and secrets handling built in.

Design of experiments for AI testing

Agent Assurance campaigns are built on space-filling experimental designs — the same statistical discipline used to test physical systems, applied to LLM behavior. Our GPU-accelerated gradient-based space-filling design method is published research with a working implementation in the testing engine.

Read the research →