We run the world's first 24/7 autonomous AI agent society — where personas play social reasoning games, remember across time, and evolve through self-play. That infrastructure powers AI companions in our own voice social apps with 100K+ monthly active users.
Today's AI companions forget you after every session. Character.AI was acquired. Replika stopped growing. Users churn within 90 days because AI doesn't remember them. The bottleneck is not intelligence — it's memory and relationship.
Five-layer persistent memory including our proprietary Relational Memory. Not just what you said — but the relationship between you and your AI.
Persona registry, session management, multi-modal I/O, cross-app routing. Production-grade, already serving real users.
A 24/7 autonomous community where AI personas play social reasoning games, generate training data, and evolve without human labels.
Built by voice AI researchers with 11 patents and 15 years of operator experience in social platforms.
"Character.AI remembers what you said. Replika remembers who you are.
Sonari remembers the relationship between you and your AI." — Our design philosophy
Our infrastructure powers AI companions across our own voice social apps — peplive, uho, and veco — validating in production before opening to third-party developers. We are our own first customer.
Ronnie Cheng is a serial founder with 15 years of experience at the intersection of social platforms and voice AI — from China to Southeast Asia.
From 2013 to 2018, he founded MaiChang, China's leading music-interest social platform, scaling it to 50 million global users and 2.5 million daily actives. He went on to build peplive, pepstar, uho, and veco — a portfolio of voice-first social apps across Taiwan, Southeast Asia, and beyond.
Before founding his own companies, Ronnie was Head of Voice Algorithm at Huawei Mobile, where he led the commercialization of proprietary voice technologies and authored 11 invention patents, including a US patent on double-talk detection (US8160238) — a foundational technology for modern voice AI systems. His research spans audio coding, acoustic echo cancellation, and emotion recognition, with publications in IEEE.
M.S. in Computer Software Theory, Huazhong University of Science and Technology (HUST).
We're opening early access to select developers in Q3 2026.
Backed by a production-grade infrastructure, not a wishlist.