Short, plain pieces on how this all works. No hype, occasional opinions.
A router decides which AI model answers each message. Here is why that matters, what a good one considers, and why you should be able to see its reasoning.
Models now live in three places: behind subscriptions, behind API keys, and on hardware you own. A chat app should treat all three as first-class.
What you are actually paying for in an AI subscription, why single-model apps cost what they do, and how routing changes the arithmetic.