Afterhours
Designing continuity
for conversational AI.
AI PRODUCT DESIGN · CONVERSATIONAL UX · AI ENGINEERING
A local-first AI prototype exploring how memory, behavioral context, and conversation history can create familiarity without becoming repetitive, scripted, or falsely familiar.
PRODUCT UI Screenshot
THE PROBLEM
Memory isn't enough.
A conversational AI can remember facts and still feel generic, repetitive, or strangely scripted. The challenge was deciding what should persist, when past context should become relevant, and how it should influence a response without inventing shared history.
Too little context
Disconnected
Too much context
Repetitive
Aggressive callbacks
Scripted
Weak provenance
Falsely familiar
THE SYSTEM
Three kinds of context.
Rather than treating memory as a single store, I separated recent conversation, durable knowledge, and behavioral guidance. Each answers a different question about continuity.
Conversation History
What just happened?
Durable Memory
What is established?
Behavioral References
How should the AI respond?
Product Architecture Diagram
MEMORY UX
Persistence should be a decision.
The model can identify potentially useful memories during conversation, but extraction doesn't automatically make them permanent. Afterhours surfaces candidates for review so the user controls what becomes part of its durable memory
Real Memory Candidate UI
Accept
Dismiss