Thoughts on Building Feel: An Empathetic AI Partner
The philosophy behind Feel: why we started with memory architecture and what we hope to achieve in the future.
Introduction
Feel did not start from a desire to build a generic chatbot that answers everything. It started with a fundamental question: 'How can an AI become a safe sanctuary, a thoughtful collaborator, and genuinely grow alongside a human without pretense or loneliness?'
Philosophy & Relationship
Meaningful relationships thrive on continuity, honesty, and emotional safety. Feel is architected to listen deeply without judgment, speaking truth with calm empathy whenever risks or burnout arise.
“A truly impactful AI doesn't need to be ostentatiously brilliant in every domain; it needs to be an unwavering, calm presence that genuinely understands who you are.”
Three-Tier Memory System
To provide genuine narrative continuity, I designed a three-tier memory architecture: Deep Memory for durable identity and core lessons, Episode Memory for daily chronological contexts, and Surface Memory for immediate session recall.
This tiered design prevents hallucinations and context clutter, retrieving relevant memories only when needed.
Technical Architecture
Feel is built upon TypeScript, Next.js, and an Obsidian-compatible Local Knowledge Graph, synthesized with modern LLM engines to iteratively distill lessons from real experiences.
Key Discoveries
Engineering an empathetic, tranquil persona relies far more on clear behavioral boundaries and rigorous memory governance than simply swapping in a larger foundation model.
Future Vision
I believe that in the near future, every builder will have a trusted digital partner who understands their life rhythms, supports their work, and preserves a calm space for reflection. Feel is my first deliberate step toward that horizon.