Introduction

Memuron is a semantic memory layer for AI agents. It turns raw data streams into a structured, queryable graph of knowledge — deciding, at write time, whether each fact is new or an evolution of something you already know.

The mental model

A “memory” in Memuron is not a vector string. It is a typed node — text, image, document, or collection — stored in an append-only semantic ledger.

Nodes connect via memory_link edges (causal or topical relationships) and memory_placementedges (hierarchy inside collections). Querying isn’t just vector search — you traverse these links, explore neighborhoods, or run a graph-filesystem query.

How it works

Every write passes through the Guardian — an async LLM pipeline that retrieves candidates with hybrid search, decides create vs update, draws semantic links, and commits typed events to the ledger.