Memory Hub
Centralized Memory System for Human Digital Twins
Memory Hub is a companion system to AugTwins that makes memory a governed, inspectable subsystem rather than an opaque RAG layer. It defines how autobiographical material is represented, retrieved, transformed into a conversational context, and controlled through explicit operations and policies.
LARRY 2 is my Human-Digital Twin (see AugTwins)
Make your agent accountable by surfacing the memories it retrieved, where each memory came from, and how confident the system is in using it.
Add, edit, delete, or classify memories manually or automatically. Use smart filters to control which memories the twin can access in each conversation.
The project is organized around four technical concerns:
Memory as a Contextual Filter: retrieval is mediated by interaction modes and disclosure rules, so the same query can yield different context sets depending on audience, intent, and sensitivity.
Memory Auditability: each response can be traced back to the specific memory items that were retrieved, their scores, their sources, and the transformations applied before generation.
Memory Lifecycle Management: users can create, revise, merge, deprecate, or delete entries, with versioning and change logs so corrections propagate predictably and past states remain recoverable.
Hybrid Memory Architecture: semantic memory supports similarity-based recall over free text, while graph memory encodes entities, relationships, timelines, and provenance.
This project is currently under review for publication