prompt drift is real but it's a model problem, not a memory problem. Open Brain works at the layer above: your context persists regardless of which model you use or how badly it handles long conversations. the models will improve. your memory shouldn't depend on them improving.