I recently developed an AI persona that takes LaTeX input into account while proofreading. I then spent considerably more time trying to configure my Discourse instance so that everyone could use it.
Long story short: In my experience, changes made to affect the AI contextual_menu often don’t take effect until you rebuild Discourse using, for example,
./launcher rebuild app
To be completely clear, I’m talking about the menu that arises when you press on the sparkly button in an editor toolbar, like so:
You might notice that, in my Discourse, the only option that pops up is “Proofread text”. Translate, Markdown table, and Smart dates don’t appear at all. The way to accomplish that is to fiddle with the “allowed groups” for the personas that are used for those various tasks. The Helper section in the AI > Features tab for my Discourse looks like so:
The Math Proofreader persona allows access by anyone at trust_level_0 (or higher, I guess). The others allow no access at all.
Here’s the thing: I tried for hours to get this to take effect and, generally, it just didn’t work. The only way that I have reliably been able to get these changes applied is to rebuild the whole instance, as I already indicated.
This has been mentioned before.
My Discourse instance is just a basic 30-minute install on a Digital Ocean droplet running Ubuntu.