The Malibu blog
Writing from the network.
Network updates, engineering notes, and the ideas behind turning every idle Apple Silicon Mac into part of the world's biggest AI cloud.
The signed receipt your inference API doesn't ship
Your inference API returns some tokens and a model field that says whatever the vendor wants it to say. There is no signed statement, from the machine that actually ran the request, tying the output you got to the model you paid for. Malibu attaches one to every response, and you can verify it in a second without trusting Malibu.
Read the postTurning on the world's biggest AI cloud
Two hundred million Apple Silicon Macs are already deployed, already paid for, and idle twenty hours a day. Together they hold more consumer AI compute than the top three hyperscalers combined. The only thing missing is a network that turns them on.
Read the post