Notes
One question per note. Claims about anyone else's product are quoted from their own pages.
A LaunchAgent with StartCalendarInterval, loaded with launchctl bootstrap. The plist, the test command, and the one behaviour that makes it better than cron on a laptop.
Eight projects worth building, with the files each one reads and how often it runs. The third column is the one that decides whether you still have it in December.
On the free plan they are hidden rather than deleted for the first year, and permanently deleted after it. Slack's own wording, and what to do before the deadline.
A vision model running on your own Mac can name a screenshot from what is in it. The commands, the prompt that makes the names usable, and what to watch out for.
Whether an upload trains a model, and how long it is kept, depends on your tier rather than the brand. One table, every row taken from the provider's own page.
Decide the columns first, run one pass over the whole folder, then reconcile the total against your card statement. The steps, and the failure modes nobody lists.
Two environment variables and a model pull. The exact setup, the context length people get wrong, and what still reaches the network afterwards.
Only the files it reads — but a file it reads becomes part of the prompt, nothing is denied by default, and what happens next depends on your tier. The precise answer.
There are three schedulers and only two of them can read the files on your Mac. The cron and LaunchAgent route, written out, with the flags an unattended run needs.
Export CSV rather than PDF, keep it in a folder on your own machine, and point something local at it. The steps, and what each AI provider does with an upload.
The zip is one folder per conversation and one JSON file per day, and the messages name people by ID. How to read it, and what the export leaves out.
Three separate things have to stay alive: the session, the Mac, and the work itself. What each one needs, and the one most guides never mention.