Thanks, kind stranger! I wrote the comment that I would have loved to find before (and after) making the leap. Stuff is changing so fast, and there's at least three tracks: "Scavenger-old-linux-box", "Fancy-AI-cube", "Mac + $$$ + RAM"

Again: I'm finding waaaay enough utility that I'm tempted to invest more "CapEx" and get a used system for day-to-day, "always on" local work... but more literally, that's probably a better job for "OpEx"! Tune my "crontab" work against local models and then max out at a $1/day budget slaved to an always on RPI connected to ethernet at home.

$365/year of off-site AI lasts 10 years before I come close to recouping the hardware (and electricity) costs of having "yet another device" purchased and turned on 24x7... and certainly there will come a day when you go to the store and buy a $200-500 "TITO" device (Tokens In => Tokens Out) that plugs into a ~30-60W USB-C port before then.

If you're using HF tokens (or "rent-a-A100" or whatever), are always connected to home ethernet (Sun Microsystems: The Network IS the Computer), and maybe supplement with a Kagi backend for attaching to the raw internet then you get _most_ of the surety of "my queries are private" unless you're locally hacked or are the target of nation-state scrutiny. :shrug:?

Keep in touch if you end up doing something cool with all this! $USERNAME@yahoo.com (and hopefully I'll have my AI setup filtering out all the viagra spam before then!).