#CompanionIntelligence EmbodiedAgents
#DeveloperUpdate 2
We are building Embodied Agents: a 3D virtual puppet using web tools and our
#GLTF avatar pipeline.
Our
#avatars offer a more immersive, engaging, and emotionally intelligent experience compared to a normal text- or voice-based agent.
Our embodied agents are more communicative. With full body poses, animated mannerisms, and subtle facial expressions, embodied agents can enhance engagement and foster a more impactful experience with users.
Embodied Agents can help in hospitality, retail, education and more.
We built a back-end private AI server built on Debian Linux, python, and node.js running on
#OpenSource models like
#Qwen3,
#Llama4,
#GPTOSS, and
#Deepseek
On personal
#privateAI server we are doing
#STT and
#TTS using VibeVoice, Wisper, and Piper
We built a front end on React, Tailwind, some custom avatar puppetry magic; one system for the face, another for the body.
VRM Avatars - that have a standard skeletons based on GLTF 3D file standards
Rendering with a mix of open source WebGL and WebGPU to render 3D.
WebSpatial is a new framework that enabled us to get in on the ground floor of the spatial web.
With the
#WebSpatial tools, we were able to easily unify and extend our existing web code for
#VisionOS,
#AndroidXR, and Meta
#HorizionOS. The next generation of spatial computing devices is right around the corner in 2026.
Shoutout to Anselm, Rahul, Bennet and Josh on the Companion Intelligence team who are making this happen! Can't wait to share what's next!