The challenge.
Users wanted to own AI characters, with memory, voice and visual identity, not just rent them from a centralised provider.
The product required multimodal inference (speech, vision, text) to run with low enough latency for real-time conversation, while settlement for ownership and usage had to happen on-chain.
Off-the-shelf stacks gave us either the AI or the Web3 half, never both with the orchestration we needed.




