Convai

Convai builds conversational AI characters that do more than talk. Its defining feature is multimodal perception: characters can see and hear their environment, recognise objects and other characters, and respond with actions as well as dialogue, so an NPC can notice you are carrying something and comment on it, or walk somewhere in response to instruction. Plugins cover Unity and Unreal with voice, memory and knowledge bases attached to each character. Two things to weigh. Perception plus action is genuinely harder than dialogue generation and is where Convai separates from text-first NPC platforms, so evaluate it on whether characters behave sensibly in your actual scene rather than on conversation quality alone. And running live inference per character has real per-session cost and latency implications, so prototype with your expected concurrency before committing to a design that assumes many talking NPCs at once.