Free has a cost structure

Pocket is free for patients, permanently. Maa is free to mothers. Pharmacy verification is free at the counter. These are not pricing decisions that can be revisited. They are load-bearing, because the people the system exists for are the people any fee would exclude first, and a health system that prices out the poorest has failed at the one thing it was for.

Free for the patient does not mean free to run. Every triage answer, every briefing, every draft read costs something to compute. When that computation is a call to a foreign frontier model, it carries a per-call fee. The fee is small. It is also multiplied by a country, every day, forever, and small times a country is not small.

This is the economic reason the model has to come home, and it is not a rounding error. A mission funded to serve tens of millions of free interactions cannot rest on a marginal cost that scales linearly with use and is set by a vendor abroad. The arithmetic does not close. Either the free promise breaks, or the cost of one more answer has to fall toward zero, and the only way it falls toward zero is a model the network runs on its own infrastructure, where the marginal cost of one more answer is electricity, not a metered call.

So the free promise and the sovereign model are the same decision seen from two sides. You cannot keep care free at national scale on rented intelligence, and you cannot justify owning the intelligence without the scale that free access creates. KhaM-Med is what lets the free promise survive contact with the bill.