Kyloen Publishes Architecture Behind Three-Layer AI Memory System for Children's Companions
Working memory, episodic key moments, and semantic pattern detection enable the most personalised AI companion experience available
New Delhi, India
Kyloen today published details of its proprietary three-layer memory architecture, the system that enables Kylo to remember a child's dreams, fears, favourite cricket players, and emotional patterns across months of interaction.
The first layer, Working Memory, maintains the last 20 messages in any conversation session, providing immediate conversational continuity. The second layer, Episodic Memory, extracts one to two memorable moments from each session and stores them in a structured key moments database. These moments are categorised as dreams, fears, achievements, phrases, patterns, friend references, conflicts, or interest signals. The third layer, Semantic Memory, computes weekly pattern analysis including mood averages by day of week, peak activity times, topics driving engagement versus disengagement, and response patterns to different interaction styles.
Why Memory Matters
In internal testing, children who experienced Kylo referencing past conversations reported significantly higher trust scores and longer average session durations compared to baseline AI interactions. When Kylo says "Last week you mentioned wanting to be a fighter pilot like your uncle," the child feels genuinely understood rather than merely processed.
"Memory is what separates a companion from a chatbot. A chatbot processes your words. A companion remembers your world," said Neha Kapoor, Founder and CEO.