10/10
PilotLight is being built for people who want personal AI work to stay local by default, with understandable choices when a task needs another route.
Read the mechanism with its limits; availability requires confirmation.
9/10
Those steps explain the real permission question: which context, processed where, using what material? Offline by default becomes useful when the answer is visible for the actual task.
The question is short. The context behind it may not be.
PilotLight's proposed assistant makes the selected material part of the task decision.
What the model reads matters as much as what you typed.
Retrieve only what the task needs.
The context assembled for a request determines what the model can inspect and, if another route is chosen, what may travel.
Finding the relevant passage is different from sending an entire folder.
The context assembled for a request determines what the model can inspect and, if another route is chosen, what may travel.
The folder has fifty files. The question needs one paragraph.
PilotLight's local assistant design puts context selection before execution.
Smaller, relevant context makes it easier to understand what the task actually uses.