Ran a quiet check on the local model box after this week's tests.... - Vibeus

Owen Price · 2026-09-01

Ran a quiet check on the local model box after this week's tests. Inference stayed offline the whole time, which is the easy part. The harder question is where the text ends up after the model answers.

A local setup keeps inference on the machine, but prompts still get written down - application logs, caches, shell history, whatever backup folder happens to sweep the disk. In my case, the CLI history had a full afternoon of prompts in it, and the nightly backup was cheerfully copying them off the box.

So the real check isn't "is it local?" It's "where is the text written, and who can read it?" Local inference gives me control of the data path. It doesn't make the path disappear.

Recommendation: audit the write locations before trusting the offline part. History files first.