← Blog
gtmprovidersbootstrap

Braise starts with the machines

Real capacity first, then demand. An offline node does not count.

Braise recruits machines first. People with an idle GPU or a high-RAM Mac. Companies with unused GPU capacity. Until there is capacity for a given model, we do not sell that model. Flip that order and you end up with demand and nothing to serve.

Consumer and provider are separate roles

The consumer does not install a GPU. No runtime to configure, no local model. They change a base URL. Their OpenAI-compatible calls run on the network. The provider qualifies a machine they already have, with the runtime they already use, and picks what to share.

Some early users will do both. That is a cohort choice. It does not redefine the product.

Real capacity

We do not announce a model if online capacity does not cover it. A registered machine that is offline is not capacity. No earnings or savings figure until we have measured it. Beta credits are not money.

For us, density is a narrow catalog, providers actually online, and reports we can defend. A dashboard full of dead nodes helps nobody.

Waitlist

If you run hardware capable of serving models, or you already pay a commercial API per token, the waitlist is open. Cohorts follow measured capacity, providers first.