Definition
Plain language
Running an AI model on your own computers instead of calling someone else's service.
As stated in the literature
Serving open-weight checkpoints on locally controlled infrastructure, removing vendor-side safety wrappers, routing, and preprocessing from the measurement path.
Also called: self-host, self-hosted, self-hostable, self-serve
Why it matters: It removes the hidden vendor-side layers that can quietly reshape a model's answers, which is essential when you need to attribute behavior to the model itself.
For example, a lab downloads an open model's weights and runs it on its own servers so that every request is handled by hardware it controls.
Heard on the show
“From sixteen-thousand they filter down to about thirty-four-hundred candidates that can actually be sandboxed — free, self-hostable, GUI-driven, doesn't require specialized hardware.”Episode 017 — When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers