Perxona Connect Kit: a web component that makes an LLM reply speak through a 3D avatar

Perxona Connect Kit is a starter kit and web component for adding a talking 3D avatar to a web app. The core idea is simple: your code produces a text reply, then calls presenter.present(reply) and the avatar speaks it with lip-sync, emotion and gestures. The repository does not ship a new language model. It ships the presentation layer around one, plus an Express sample, a Studio flow for choosing avatars and voices, and a tool for uploading your own .vrm character. The README frames this as a five-minute integration. That timeline covers getting two API keys from console.perxona.ai, running the sample locally on Node 22+ or deploying it to Vercel, and picking an avatar, scene and voice in Studio. It is a setup estimate from the project documentation, not an independently measured benchmark, and it assumes you already have an agent or are willing to use the hosted Perxona chatbot. One element, two keys Integration is centered on a single web component element. Your server is expected to tell the browser which presenter engine to load and hand it a publishable key, while the secret key stays server-side. The sample implements this with GET /api/config and GET /api/connect-key. If you already have a backend — the README lists OpenAI, Claude, LangChain or a custom service as examples — you keep that logic and pass its output text to the avatar. Because it is a web component, the same code is intended to run in a standard page, an Electron app, a Chrome extension or a kiosk browser, starting from an embed demo of about 170 lines. The kit also includes