Aegis Cognition · v0.3.0
Voice AI Companion · Ages 4–7

The First AI Friend
Built for Children

A COPPA-compliant, bilingual voice companion grounded in Harvard's Serve & Return framework. Not a toy. Not a chatbot. A cognitive environment.

105/105
Safety Tests Passed
<800ms
Voice Latency Target
3
Safety Layers
2
Languages RU · EN
🛡
Aegis Safety
3-layer protection: regex, LLM euphemism detection, output validation. COPPA & GDPR compliant.
🧠
Harvard S&R
Every exchange follows the Serve & Return neuropsychological model. 2s cognitive pause built in.
🎯
Vygotsky ZPD
Responses adapt to the child's Zone of Proximal Development — always one step ahead, never overwhelming.
Zero Latency
Dual-queue streaming architecture. TTS starts per sentence, not after full LLM response.
🔒
Zero-Knowledge
90% of child data stays on-device. Cloud receives only anonymous semantic vectors.
Claude 3.5 Sonnet
FastAPI
ElevenLabs TTS
Groq Whisper
WebSocket PCM
Three.js
COPPA Compliant
Bilingual RU·EN