MODEL-OS · MAPPING 04 / 05
Filesystem Vector store
FILESYSTEM ⇢ VECTOR STORE
Paths retrieve locations; vectors retrieve meaning.
A filesystem is a tree with paths for coordinates: inodes, directory entries, byte-precise addressing. It demands that you remember where things are — and when you don't, you dig with find and grep. The machine stores everything and understands nothing.
A vector store uses meaning as its coordinate. A conversation, a figure, a failed experiment — all embedded in one semantic space, recalled by similarity. Memory becomes layered: working memory in context, episodic memory in vectors, long-term knowledge in weights. Most importantly, forgetting becomes a feature instead of a fault: unimportant memories decay along a forgetting curve.
Filesystem
- Directory tree
- Path addressing
- inodes / block devices
- Explicit deletion
Vector store
- Semantic space
- Meaning addressing
- Embeddings / similarity
- Forgetting-curve GC
LIVE · THE SAME JOB, SAID TWO WAYS
Meaning addressing
The retrieval key becomes a semantic vector — finding it no longer depends on remembering where it lives.
Layered memory
Working memory (context), episodic memory (vectors), long-term knowledge (weights) — tiered by semantics, not speed.
Forgetting is a feature
GC becomes a forgetting curve: low-value memories decay; important ones consolidate through recall.