One service, one store, three doors in.
MemState is deliberately simple to operate: a single process hosts the APIs, talks to one embedded graph database, and schedules background maintenance after responses return. That keeps latency predictable and deployments small.
What this architecture buys you
-
Immediate consistency — every surface reads the same store; no replication lag between a vector index and a graph.
-
Low operations burden — no separate database fleet to patch and scale for the prototype footprint.
-
Clear failure domain — when something breaks, the boundary is the service you already deploy.
How teams adopt it
Start with a single instance beside your agent. Standardize the observation envelope and context-question format once; MemState keeps revise and forget inside the service as you scale. Grow into HA or externalized storage when traffic and compliance require it.