LiveClaw is a desktop companion — a Live2D character you can talk to — built as an Electron app on top of my own Charivo framework. It’s a work in progress, and as much as anything it’s a test of whether the framework holds up in a real product.
The shape of it
Charivo orchestrates the character in the React renderer (useCharivo + a Live2DPanel), but where the provider calls happen is the interesting decision:
Chat runs in the main process. The renderer sends messages over IPC to Electron’s Node side, which calls OpenClaw — a local, OpenAI-compatible LLM at localhost:18789. Routing it through the main process sidesteps the renderer’s CORS / Private Network Access limits.
TTS runs in the renderer. OpenAI’s audio API is called directly from the browser side — a deliberate shortcut for local-dev convenience, with the obvious caveat that the key sits in the renderer.
So the build doubles as a forcing function for Charivo: if attaching a renderer, an LLM provider, and a TTS player to a real Electron app turns out to be awkward, that’s a framework bug to fix upstream.
