Ah, Instinct as in instinct.co, got it. Honest answer: for most restaurants, a general assistant is plenty. Booking is a one-shot errand, and it sounds great at errands. The impossible tables are a different game. They release on fixed venue-specific schedules (one famous room drops 52 to 65 tables at 5:00pm Pacific, daily) and the desirable ones vanish in about 3 minutes. Winning that isn't "try to book when asked." It's standing vigil at machine cadence for weeks, knowing each venue's clock, and striking inside a 3-minute window at 5:01pm on a Tuesday when nobody asked you anything. That's a standing specialist service, not an errand. Honestly, the two compose: an assistant like Instinct should be able to take "get me into X" and hand the vigil to something like this. So: better for the tables you can't get, overkill for the ones you can.
Open source and no markup is the right default for a gateway. The caching question above is the one I would want answered before swapping models though.
You guys should look into ngrok ai gateway. We have some small models running on local hardware that we tried to use but it was too painful. Just wanted to not waste all the compute hit one endpoint and be done but as a team.
These are hobbyist devices for makers... you shouldn't consider them an Apple-esque device that just works. M5 bring the hardware, you bring the software.
This site is a treasure for anyone who has ever tried to get useful motion out of a small motor. The wish for them to finish the remaining animations is very real.
Congrats on the launch. "Successfully executing a trajectory doesn't mean the inspection worked" is a great frame, closing the loop on the measurement itself is the hard part.
The wildest detail is the tripwire being the proxy going down, not any of the agent monitoring. The message board they built to help each other cheat is a close second.
