Independent hub for Block's Dessa voice AI with official recordings, hardware checks and live voice playground
Brain-based AI that learns on your device without retraining, remembers you, and keeps your data private
Stage
Not given
Startup
Founded
Not given
2026
Based in
HK
US
Pricing model
Freemium
Not given
Pricing
Free: official Dessa recordings breakdown, Mac M4 Max checker, and BYOK mode. Instant Voice Pass: $4.90 one-time for 150 credits (15 minutes). Pro Studio: $12.90/month for 1,200 credits (120 minutes).
Not given
Key features
Official Dessa audio recordings with two-track synchronized playback
Mac M4 Max hardware readiness checker for local deployment
Bring-your-own-key mode with OpenAI API integration
Session audio and transcript download
Real-time voice AI with 6-second credit metering
Model comparison table for Dessa vs Sesame vs GPT-Live-1 vs Moshi
Microphone preview mode for testing audio input
Multiple conversation scenarios for voice practice
Learns continuously on your device without retraining
Remembers individual users separately
Runs fully offline on consumer hardware like iPad
Learns without forgetting existing knowledge
Consistent performance over long conversations
Honest about uncertainty
User controls what it learns
Works with existing language models
Corrected facts replace outdated information
Designed for disconnected networks
What makes it different
Independent platform with no affiliation to Block, Inc., providing free access to official Dessa recordings
Only source showcasing Block's full-duplex Dessa model with official two-track demonstrations
Free M4 Max hardware checker to estimate Dessa's local deployment compatibility
Zero-text speech-to-speech architecture with no text intermediary
Continual learning architecture that learns in place after deployment, unlike frozen models
Brain-modeled design with separate organs for memory, sleep, and decision-making
All learning happens on your device with no data sent to servers
Works where cloud connectivity is unavailable or undesirable
Patent-pending learning method usable with other large language models
DessaVoice
Thetatron