That’s a pretty slick way to use hardware that would otherwise just sit there sipping power and judging you. I like the choice of keeping the two Ollama boxes independent instead of trying to make one “cluster brain” out of them, way less fragile and a lot easier to troubleshoot when something acts dumb at 11 PM. The dispatcher idea is the neat part here, especially with Wake-on-LAN baked in. For Obsidian, I’d probably lean toward Open WebUI Knowledge/RAG first, just because it keeps Raymond as the one front door instead of turning every app into its own special snowflake integration.
