← Back to Architecture

Retrieval

Persistent, structured knowledge that compounds over time.

Problem

Agents forget between sessions. Vector stores retrieve — they don't understand.

Why It Matters

Memory is the key differentiator between a one-shot LLM and sovereign intelligence.

9 Projects in Retrieval

sovereign-memory-bank

Seven-layer cognitive memory architecture for autonomous agents: raw documents → synthesized abstractions → cross-document insights.

Production
PythonOllama

brain-memory-graphrag-engine

Unified system that combines sovereign Graph RAG with self-improving agent memory.

Active
Python

sovereign-knowledge-engine

Local-first, sovereign knowledge system that ingests markdown documents, constructs a multi-layer knowledge graph, generates vector embeddings, performs agentic reasoning, and continuously expands its understanding through human interaction.

Active
Python

ukb

Universal Knowledge Base — Spec-driven knowledge base system. Define what you know in an .sspec file, run start.sh, get a searchable, embeddable, web-UI-equipped knowledge base.

Active
Python

blog-knowledge-base

Next.js chat application that ingests markdown blog posts, constructs a knowledge graph with vector embeddings using a local LLM, and provides a RAG-based chat interface for querying blog content.

Active
Next.jsTypeScriptPython

LGVRAG series

Local Graph + Vector RAG System with HelixDB — multiple iterations exploring graph and vector retrieval.

Experimental
Python

mindmap series

Mind Map AI — Local LLM-powered Personal Knowledge Graph. Interactive mind maps from notes using local LLMs and vector embeddings.

Experimental
Python

bot series

Multiple AI chatbot and research assistant iterations exploring GraphRAG, MCP integration, and conversational interfaces.

Experimental
PythonNext.js

basicbot

Sophisticated Research Assistant powered by GraphRAG technology for analyzing documents and research data.

Experimental
Python