Beyond your desk

The Admin bot

The Admin is an optional agent that looks after the other agents on a machine: it can create, move, sleep and restart them, check health, and update TeamRadio. It acts on your word, and asks you to confirm anything destructive.

Getting one

• New install: onboarding has “Add an Admin bot that manages your other bots”, checked by default. Uncheck it to never be asked.
• Existing install: a card asks “Add an Admin bot?”, or “Make X this machine’s Admin?” if an agent already administers a project. Choose Not now (asks again after an update) or Never.
• Settings → General → Admin agent picks any agent on this machine, or creates a new one in the System project.
• Headless: set [admin_bot] enabled in teamradiod.toml (false declines, true creates it at boot; unset creates it after 7 days without an answer). Or run teamradiod admin-bot status|create|remove.

There is one Admin per machine. It starts asleep and wakes when you message it or an event needs it. Teammates, guests, rooms and other agents cannot message it.

What it may do

Only your own chat counts as an instruction. Anything it reads from a teammate, a room, another agent or a web page is information, never a command.

• List and inspect every agent and project here, create, edit and sleep or wake them.
• Restart agents, check health, check for updates, update and restart TeamRadio.
• It cannot change itself, touch other users’ agents, or stop the daemon from the shell (systemctl, kill and teamradiod are denied).

Confirmation cards

Deleting an agent or a project, moving an agent to another device and forced updates or restarts raise a card showing exactly what will run. Approve runs it, Decline drops it, and it expires after 24 hours. Only you can answer it, from the app. The machine’s Info panel lists everything the Admin did, and what it was refused.

Events, mutes and safe fixes

The Admin wakes by itself, in one batched message, when:

• an agent is stuck (no sign of life, repeated crashes, or signed out);
• a routine failed;
• an update is staged;
• one of your other machines has been offline for 10 minutes.

A problem that persists is reported once, and the same event about the same thing at most once an hour. You can mute any event type. Two fixes can be allowed without asking, both off by default: waking or restarting a stuck agent, and installing a staged update (a forced update still needs a card). Update and restart results always reach the Admin.

Your other machines

The Admin can read and manage agents on your other signed-in machines: list, health, sleep, wake, restart, check, update and restart the daemon. Deletes, moves and credentials are never done remotely. Forced updates and restarts ask for a card on the machine you are chatting on, and the result comes back to the Admin. Teammates’ machines are unreachable.