1What it is

The Member Management AI is an assistant dedicated to the founder and admins. It doesn’t answer members and doesn’t explain how Tavora works: it works on the members of your organization — searching, filtering, counting them, and acting on your request: tags, roles, invites, removals, bans, messages.

It’s separate from Tavora AI (the assistant that answers members’ questions about the Library materials): a different AI, a different job, a different memory. And like everything in Tavora it’s isolated per community: it sees only your organization, never the others.

The fastest way to get it: write what you would do by hand. “Who has the Volunteer tag but isn’t in the Events group?” · “Give the Senior role to everyone who joined 6 months ago” · “Send a reminder to the coordinators”. It does the work, you confirm.

2Where to find it

Three ways in, same AI:

From the community menu → Members → 🤖 Member Management AI: the panel opens in private chat, with a button to make a request and your saved scheduled instructions.
In linked groups: the /aistaff command followed by your request (e.g. /aistaff how many are we?). It replies in the group; if the answer contains members’ private data, it delivers it to you privately.
In private chat: the same /aistaff command with your request, without going through the menu.

Only the founder and admins can use it: for anyone else the command does nothing more than note it’s staff-only.

3What it can do

Look things up

Lists and filters: by role, custom role, tag, group, seniority, activity (proposals created, tasks completed). Filters combine: “Volunteers who joined over 6 months ago and aren’t in the Rome group”.
A member’s card: roles, tags, groups they’ve been seen in, active sanctions, activity.
Organization statistics and the audit log of recent management actions (who did what, when).
Export of the list as CSV or TXT, with the same filters as your request.

Act on members

Assign and remove tags and custom roles.
Invite to a group (single-use link valid 24 hours), remove from a group, ban and unban from the organization. A ban comes in the same two kinds as on the member card: by default from the organization only (the person stays in the Telegram groups) or, if you ask to remove them from the groups too, a full ban from the organization and every linked group.
Send private messages to members (with a preview of the text before sending) and post announcements in linked groups.
Write internal notes on a member’s card, visible to staff only, and set or remove their custom fields, one member at a time.

Organize the structure

Create, rename, reorder and delete tags and custom roles.
Set a role’s permissions, or the base permissions shared by all members.

Feel free to compose multi-step requests: “create the Veteran tag, assign it to everyone who’s been with us for a year and notify them privately” — the AI completes them all, in the right order.

4How it executes: you ask, you confirm

The rule is designed to be fast without ever losing control:

A clear request about 1 person → it executes right away and reports the real outcome.
2 or more recipients, a group announcement or a message to members → it prepares everything and shows you the preview. Nothing goes out until you reply “confirm”.
Several actions in the same request → a single confirmation: one “confirm” executes them all in order (up to 8), and the AI reports the outcome of each.

A confirmation only counts if it comes, in a message after the preview, from whoever made the request: nobody else can confirm for you. The proposal stays confirmable for one hour and only in its own organization: after that the bot tells you it has expired, and a “go ahead” no longer runs it, not even after 💬 Continue. And you can always reply “cancel” to stop everything.

Change your mind? Ask it to cancel or correct: the AI rebuilds the updated proposal. And the outcome is always honest: it tells you what was actually executed, what wasn’t, and why.

5Scheduled instructions

From the panel → ⚙️ Scheduled instructions you can save up to 5 rules the AI applies on its own, on the days and at the time you choose, so you don’t have to remember:

“After 30 days of seniority, assign the Active member role”.
“Whoever passes 6 months gets the Veteran tag and a congratulations message”.

Each rule has its own 🔄 Schedule (every day, the weekdays you pick in 📆 Day, or every N days) and its own 🕐 Time, in the organization’s time zone. If the text already names a day or a time (“from Friday to Sunday at 8 pm”), the rule is set up that way and the bot tells you as soon as you save it. A rule’s text can be up to 1000 characters.

Automatically, without asking you, the AI only applies tags, custom roles, private messages the rule calls for, unbans, group invites and internal notes. Removals, bans, warnings, recognitions and changes to custom fields never run on their own: they become suggestions to confirm. Group announcements, changes to the tag and role catalog or to permissions, and the visibility of card fields are not available automatically: only in a direct conversation.

The report reaches you in private only when something happened, in the organization’s language: what was applied and the suggestions, with people’s names and the reason. You confirm the suggestions with ✅ Confirm or drop them with ❌ Cancel, within 24 hours and only with those buttons: a “confirm” typed in the panel doesn’t run them, and a new request in the panel doesn’t cancel them. One ✅ covers at most 8 suggestions, and the report says which ones.

In the panel, the list shows each rule’s schedule, next run and who created it. Tap the number to open its card, with the last run and the buttons ✏️ Edit text, ⏸ Pause / ▶️ Resume and 🗑 Delete, which asks for confirmation with 🗑 Yes, delete.

6In groups

With /aistaff in a linked group the answer lands in the group, handy for quick things: “how many are we?”, “how did last week go?”.

But a group is a public place: whenever the AI has to read members’ private data to answer — member cards, member lists, the groups someone has been seen in, recent staff actions, the banned list — the answer always arrives in private, and the group only gets “🤖 I replied to you in private”. Previews awaiting confirmation and export files always arrive in private.

In the group the AI writes in the organization’s language, doesn’t use the memory of your private chat, and only knows the Library documents the group can read.

7Safety and limits

Maximum 20 recipients per action: mass actions (“ban everyone”) are refused.
Never against the staff: it doesn’t act on the founder or admins, and can’t remove itself. The only exceptions are tags and roles an admin requests for themselves.
Removals and bans: on one person, with a clear request, it runs them right away like any other action; on several people they go through your written confirmation; in scheduled instructions they are only suggested, never executed.
Total isolation: it sees and touches only your community. Banned members receive no message from the bot.
A daily request cap per community, against excessive use.
Everything tracked: every action lands in the organization’s audit log, signed “via AI”.

The AI can’t be manipulated: instructions only count when they come from staff in chat, not when hidden inside data (say, a username). And it refuses to compose offensive, deceptive or discriminatory content — always offering you a correct alternative.

8The two AIs compared

Two assistants with different jobs live in Tavora:

Tavora AI — for all members: answers questions about the Library materials and the community. It touches nothing: it informs.
Member Management AI — for founder and admins: consults member data and executes management actions: right away on one person, with written confirmation on several people or for messages and announcements, and within limits. It’s an operational tool.

Separate memories, separate permissions, no communication between the two: what you ask the staff AI stays with the staff.