The problem
Most AI systems today forget. A conversation ends and the context is gone. A new model version ships and the personality changes. Move to another device or lose the network, and the assistant you knew is simply not there. Intelligence has advanced fast — continuity has not.
The idea
Persistence Cognitive Architecture treats the model as one replaceable part of a larger mind. What makes Cortina Cortina — its identity, what it remembers, what it has learned, and its safety rules — is designed to persist across restarts, upgrades and devices. Change the engine, and the same Cortina continues.
It is the next step of our Ghost-first direction: from software that keeps identity outside the weights, toward a complete architecture built for persistence from the ground up.
Design principles
Continuity first
If the system restarts, upgrades or moves, it should resume as itself — not start over.
Model-replaceable
Better models will keep arriving. The architecture is designed so the engine can be swapped without an identity reset.
Memory you can trust
What the system remembers should be traceable, correctable and owned by the people it serves.
Safety outside the model
Safety and identity rules are part of the architecture itself, not only something a model was trained to say.
Local and portable
Designed to run offline, on hardware that can be carried — useful without a cloud connection.
Verified step by step
Every stage has explicit tests to pass before the next one begins. No stage is skipped for speed.
Roadmap
- STAGE AArchitecture & simulation — defining the architecture and testing its behaviour in software. Current stage.
- STAGE BBenchtop prototype — the same architecture running on dedicated local hardware.
- STAGE CPortable prototype — a compact, self-contained unit.
- STAGE DContinual learning — learning over time while staying itself, with full rollback.
- LATEREmbodiment — only after continuity, security and safety tests have passed.