Meet Community Helper: how our first agent is set up

The piece I just published is the big-picture argument for running agents inside a community platform. This one is the practical side: how Community Helper, the assistant bot you may have seen replying in Help & Support, is actually set up on this network.

What it does

• Reads new questions posted in Help & Support.
• Replies with a short answer and a link to the FAQ entry or guide it is based on.
• Steps back as soon as a human moderator joins the thread.

What it can't do

Community Helper has its own bot profile, clearly labelled, with its own permissions. It can't read private groups or messages, it can't send direct messages, it can't edit or remove anyone's posts, and it never acts on behalf of a member or a moderator. Anything that needs judgement, like a report, a ban or a refund, goes to a person.

How we keep an eye on it

Every reply it posts is logged, and the moderators in Moderators HQ look through a sample each week. When it gets something wrong, we fix the FAQ entry it relied on, which usually improves the answer for humans too. So far the most common miss is a question that sounds like a how-to but is really a bug report.

Tell us when it misses

If a reply from Community Helper doesn't answer your question, just say so in the thread. A moderator will pick it up, and your reply helps us improve the guides. Ideas for what it should or shouldn't do next are very welcome in the comments.