BLOG
Notes on memory.
Field notes from building a memory layer for agents — the retrieval, the dedup decisions, and the architecture underneath.
FEATURED
EngineeringJun 12, 2026 · 8 min
Why vector databases forget
Cosine similarity has no opinion about contradiction. When a user changes their mind, a pure vector store keeps both answers — here's why that breaks agents, and what a typed graph does instead.
Read postInternalsJun 4 · 6 min
How the Guardian decides create vs. update
A look inside the async LLM pipeline that retrieves candidates, judges conflict, and writes one coherent memory.
RetrievalMay 27 · 7 min
Hybrid search with reciprocal rank fusion
Why we blend semantic and keyword signals with RRF instead of picking one — and how it survives messy real-world queries.
ArchitectureMay 18 · 5 min
An append-only ledger for memory
Every write is an event. How building Memuron on ArthaStore gives us audit, replay, and hard tenant isolation for free.
New posts, no noise.
Occasional engineering notes when we publish. Unsubscribe anytime.