Self-Hosted AI

Why I Built a Priority Queue for Ollama

Sixteen AI agents sharing one GPU. The GPU was fast enough. Ollama's FIFO queue wasn't. Here's how 620 lines of Node.js fixed it.