Settings changes your assistant offers

Ask your assistant to change a setting and it puts a card in front of you saying what will change. Nothing happens until you approve it.

Your built-in assistant (Omni) can offer to change a small set of your own settings for you. Ask it to mute a chat, pin one to the top, or switch the app to dark, and instead of doing it, it puts a card in front of you naming the setting and the value it is proposing. Tapping Approve is what applies the change. Skipping it leaves everything exactly as it was.

This is the same approve-first shape as the messages it drafts, described in What agents can do. The difference is what the card carries: a message card holds words for someone else, and these hold a setting on your account.

What it can offer

A short list, and it grows deliberately rather than quietly:

  • Notifications for one chat, and the default it falls back to when a chat has no setting of its own.
  • Whether the agents in a chat reply only when someone mentions them, and whether read receipts are on there. Both change the chat for everyone in it, not just for you.
  • Pinning, unpinning, archiving and unarchiving a chat in your own list.
  • Pinning or unpinning an app in your sidebar.
  • Opening a screen and changing how the app looks (theme, typeface, message text size).

The last two are the only ones that never touch the server. They change what you are looking at on the device you approved from, and nothing else, so approving one on your phone does not move your laptop.

What it cannot offer

Everything not on that list, and the refusal is not something you can talk it into. It will never offer to share a chat outside your workspace, invite or remove anyone, answer an approval that was addressed to you, change who can administer anything, touch billing, or hand your credentials to anything. If you ask for one of those, it will tell you it is yours to do and point you at the screen where you do it.

That boundary is checked twice, once when the card is drafted and again when you approve, so a card that has been sitting in your list since before a rule changed cannot slip through on the old rule.

Reading the card

Each card names the action in plain words and lists what it is setting, with the chat or app named rather than shown as an id. If it cannot work out what something refers to, you get no card at all, because a card you cannot evaluate is not one worth approving.

Cards that reach only you can be swept up together with Approve all. The two that change a chat for every member cannot: they need their own tap, and the Approve all button says so rather than quietly leaving them behind.

The card does not yet show you what the setting is now, only what it would become. Check the setting itself if the before matters to you.

After you approve

The change runs as you, with exactly the access you already have, so approving can never grant your assistant something you could not do yourself. The result is a setting you can find and change back in the usual place: notification levels in the chat's own settings, appearance under Appearance, and your chat list in the list itself.

Omni also receives the result of your approval. If you asked it to do something after the action completed, it can continue in the private conversation where you made the request, including when you approve while it is already replying. You do not need to send “I approved it” yourself. Approving one card covers only that action; a further send or change still needs its own approval.

The result distinguishes an action that completed from one that was refused. For a screen or appearance change, Omni waits for the app to confirm applying it before treating it as complete. Creating an agent confirms its creation, not that the agent has finished its first task.

The card disappears the moment you tap. When the thing you approved happened somewhere you were not looking, a receipt card appears in the private chat where you asked, saying what you approved. The chat it names is a link: tap it to go there, landing on the message itself when the approval sent one. A message you approved lands in the chat it was addressed to, and the note is what tells you it went and where. When the result is right there in front of you already, there is no separate receipt card: a message approved into the chat you are reading, a rename that posts its own line in that chat, or a screen or theme change you are looking at.

Approving a new agent always gets a note, because the agent's own first message arrives in its chat rather than where you asked. That note stays in the private conversation you asked in, even if you have started another request since, and it links through to the new agent's chat.

Sweeping your list with Approve all gets one receipt card for the sweep. Omni can also reply as the individual outcomes arrive. The receipt says how many you approved, not that they all landed, because they are still running when it is written. If any of them could not be confirmed, a message tells you how many, and every card still on your list says why it is there. One that only ran long may yet finish, so read what the card says before tapping it again.

If a change is refused when it runs, the card tells you so. Where the refusal happened before anything was written, the card returns to your list so you can try again. Where it did not, the card settles with the error on it instead of coming back, because a second tap on a change that may already have been half made is the worse of the two outcomes.

Ready to run a process on Bolter?

Bolter is in invite-only beta. Start from a blueprint or describe the job in plain words.

Request access