Abstract
A personal memory system for AI agents keeps one core graph of memories. Around it sits one private subgraph, called peripheral memory, for each installed module: a plug-in, skill, tool or device. Each subgraph lives in its own storage container, named by a whitelisted identifier, and is loaded lazily through a single hydration function. That function loads the container on first touch and refuses to create an empty subgraph over stored data it cannot read. Every write is recorded in a per-namespace journal before it is acknowledged; the journal is replayed on first touch and deleted only after a successful checkpoint. A persistent marker can request a checkpoint after every write. Links between core and peripheral nodes are kept in a separate bridge store that survives unloading. Usage telemetry per peripheral pattern drives a fluency ladder, and every sleep cycle decays idle telemetry and checkpoints all peripheral state. The document gives the data model, the steps, pseudocode, variants and the closest prior work.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Prokhorov, Artem, "Peripheral Memory: Lazily Hydrated Per-Module Subgraphs Around a Core Memory Graph", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12683