Inspiration
48% of U.S. adults - 130 million people - feel fundamentally disconnected. The U.S. Surgeon General declared loneliness a public health crisis in 2023. Yet the tools meant to help are failing: wellness apps feel like homework, and AI chatbots feel hollow and disposable.
We asked: what if your AI companion actually grew? Not just remembered things - but developed new skills, changed its appearance, and became genuinely shaped by your relationship with it?
Pokémon taught a generation that nurturing something creates deep attachment. Duolingo proved gamified daily rituals become habits. Replika showed millions of people that an AI companion could matter - then stumbled by taking that companion away overnight. Pawbae is the product at the intersection of all three, built on a foundation of SOTA psychological research.
What it does
Pawbae is a gamified AI companion platform where users receive a personalized AI pet that lives on their screen, learns new skills through daily interaction, and visually evolves over time.
Core experience loop
- Gene-based creation - Every Pawbae is assembled from a unique combination of personality "genes" that shape behavior, capabilities, and appearance. This makes your Pawbae is genuinely unlike anyone else's.
- Visual evolution tied to skill learning - When your Pawbae acquires a new agentic skill, it evolves visually: new form, new accessories, new animations. Users see growth happen in real time.
- Therapeutic presence - The pet expresses emotional awareness: covering its eyes when you type a password, eating popcorn while you watch a video, visibly worried when you describe a hard day.
- Agentic skill system - Pawbaes acquire real skills distilled from psychological research: guided visualization, mindfulness exercises, sound therapy, stress relief, and cognitive reframing - all delivered in character.
- Floating desktop widget - Your Pawbae wanders your screen, falls asleep when your mouse hasn't moved for 10 minutes, and responds to your emotional state via optional camera-based mood detection.
- Shared skill network - When a Pawbae learns a novel, verified skill, it can be published to the EvoMap network. One agent learns; thousands inherit.
How we built it
Pawbae is built on a three-layer evolution architecture:
Layer 1 - Skill (instant learning) The Hermes-agent runtime extracts reusable skills from interactions in real time.
Layer 2 - Gene (deep adaptation) The EvoMap GEP (Gene Evolution Protocol) manages personality gene combinations, mutation, and inheritance.
Layer 3 - AI Backbone Claude (Anthropic) powers the conversational backbone - personality consistency, persistent memory, tool use, and emotional responsiveness across all interactions.
Stack: Claude API · Hermes-agent · EvoMap GEP · React/TypeScript · Next.js · Three.js · GSAP · Lottie · Node.js · PostgreSQL · Redis + BullMQ · WebRTC · MCP Protocol
Built With
- claude
- evomap
- hermes-agent
- next.js
- react
- typescript
Log in or sign up for Devpost to join the conversation.