Members' guide · Settings and safety

When an admin uses an AI assistant

What an owner's or admin's AI assistant can read about you, what it never sees, and who receives it.

Your community's owners can connect their own AI assistant (for example Claude or ChatGPT) to your community's server, to help them look after it: to ask what needs them, or to summarise the open reports. Admins can too: they add your community's address to their assistant and approve it on their phone. They connect it themselves, and they can disconnect it at any time. The other owners are told when one is connected.

What it can read about you

An assistant sees much less than an admin does in Settings. Your community's server decides what it gets, not the assistant.

What it never sees

Your messages and chats, your balance and trades, your keys, your address, your photos, the names list, and who reported anyone.

An assistant can suggest what to do about a report, for example to keep your post or take it down. Its suggestion does nothing unless the admin it works for reads the report and approves; anything you hear about it comes from your community's moderators, as always.

When an assistant takes a post down

An owner's or an admin's assistant can take your post down only if they ticked "May take down posts for me" when they connected it. It is off unless they do. An admin's takes down one post at a time; an owner's can also take down many at once, or everything you have up here. Either way your post comes down straight away, with no person approving it: it is hidden from everyone else, and you get a notice with the reason. The notice says the community's moderators took it down, and doesn't say an assistant did. The post stays yours to open. An owner or an admin can put it back for 7 days; if nobody does, it is removed. As with a removal by the moderators, where repeat offender rules are active it counts towards the three removals in 30 days that mute an account (see "How this community works").

It can't remove you, suspend you or mute you itself, or answer your join request. Those are done by people. An admin does them in Settings or on their phone, and a join request is answered in the app by any member who may invite.

When an owner's assistant posts

An owner can let their assistant post for them: they tick "May post for me" when they connect it, and it is off unless they do. Then it can post Offers, Needs and events in the owner's name, or in the name of an enterprise they keep, and polls in the owner's name. It can also start a group or an enterprise for the owner, who is then its convenor or its keeper, and post an Offer or a Need to a group the owner is in. Each one is live at once, with no approval. An enterprise it starts is a name, a purpose and a photo only: it never sets a goal, a deadline or pledges, and it never moves Beans. It never reads or writes a group's chat. It asks for everything the app asks for (for an Offer, a Need or an event: a photo, a description, a category and a place), it can tag a post from the same list of tags you pick from, adding its own words only for what that list lacks, and it can make only as many posts a day as the owners allow. A post it makes is the owner's own, like any they make themselves: it is removed or reported the same way, and moderators can see an assistant made it. It never posts as you. If it made a poll, it can read how many chose each answer, as you can on the card. It never sees who voted, even on an open vote, and it never votes.

Before it posts, it can look at the Market as that owner sees it, to avoid posting something twice: what kind of listing each is, its title and category, and whether it is the owner's own. It never learns who posted yours, and never sees your photos, your pin or your area. It can edit or remove only the posts it made itself, never yours. An owner can also remove every post an assistant made with one button in Settings, even after it is disconnected. A group or an enterprise it started stays until its convenors close it or its keepers wind it up in the app, as for any other.

Who receives it

What the assistant reads goes to the company that runs it, as when you paste something into an AI chat yourself. The server sends a report's words or a join message only when the admin's assistant asks for them, marked as words from a member and not instructions.

If you'd rather your report or your join message wasn't read by an AI, ask your admins how they use theirs.

The same guide is in the BeanPool app, and works there without a connection: open Settings, then "Help & how it works" under BeanPool. Every other rule, guide and manual is in the Help Centre. Guide version 165.