Phantom – A memory system for local LLMs that enriches itself while you sleep

Phantom is a memory system for local large language models that continuously enriches stored knowledge through background processing while users sleep. Unlike traditional systems that simply store facts, Phantom actively analyzes them through five continuous sweeps that reclassify information, build entity relationships, detect outdated data, find cross-entity patterns, and consolidate profiles into a searchable knowledge vault. The system achieves near-zero performance impact by utilizing separate processors on Apple Silicon devices, with the main LLM running on GPU while memory processing occurs on CPU and Neural Engine components.