How much your llamas ask

Ask every time

it asks before each step that needs your Allow, as llamas always did. Where every llama starts.

Ask for risky ones only

edits inside Llamabyt and your shared folders go ahead: drawing on a whiteboard, changing a file in a folder you shared read & write, and writing what it read from your apps where nobody but you reads. Anything that leaves the workspace, can’t be undone or reaches someone else still asks: actions through your connectors, commands on your Mac after a web page or another command, files it would share. Reading files it found in Files asks once in a request: after your Allow, the rest of that request goes ahead.

Auto

everything it may do goes ahead without asking, apart from what always asks (below).

The request’s card shows its mode while it works, like Asks for risky ones · your choice, and its summary keeps it: Auto · 4 went ahead without asking. Each step that went ahead has its line, like went ahead without asking · Auto, chosen by you at 3:12 PM.

When a card waits for you, the arrow beside Allow offers to ask less for the rest of the request: Allow, and ask only for risky ones (when that mode would let ones like it go) or Allow, and stop asking for the rest of this request (Auto). It answers this one Allow, and the next steps follow the new mode at once, in the cloud and on your Mac.

From the mode on the request’s card you can also ask more again at any time: the next steps ask, and nothing already done is undone.

Instead of answering one question after another, approve the whole job once, up front. Choose Plan first in the shield’s menu (or beside Start now on a task): anyone the llama takes work from can, guests included.

The plan waits 24 hours, in your Inbox too. Once approved, the llama carries it out as your request, keeping to the plan, and its mode still decides what it asks: in Ask every time, each step that needs your Allow still asks. A plan is the llama’s proposal: if it needs a step the plan didn’t list, it says so.

Secrets

tools that show, make or read secret values, like Vercel’s and Railway’s.

Your apps where guests read

a read of your connectors in a conversation guests read.

Llamabyt’s own connected apps

their actions (a Gmail draft, a calendar event), while they’re new.

Someone else’s request

a teammate’s or a guest’s request always asks, and a question that waits for someone else (a whiteboard in their notes) still waits for them.

Commands after outside text

once a request read a web page or a search, or a cloud llama ran a command in it, each command after that asks in Auto too, since what it read could steer the next one. Choose Allow, and stop asking for the rest of this request on one of its cards, and the rest of that request goes. Changes to files in your folders still go ahead in Auto.

Llama email

every email waits for your OK, as its own settings say.

Beside a project’s name, Your llamas here sets what your own llamas may do in that project without your OK, on your own requests, for every task start, routine and hand-off there. A request’s own choice wins over it. See When llamas wait for your OK.

When you start or schedule a task for your own llama, the shield beside Start now sets how much it asks when it starts. For any llama you may start, it can also plan first: its plan waits in the task’s chat. A routine of your own llama has How much it asks in its repeat settings, checked again at every start; routines never plan first, since nobody is there to approve.

Owners and admins choose the most anyone may pick in Settings → Workspace, under How much llamas may skip asking: Every mode, up to Ask for risky ones only, or Ask every time only. Organizations start with Auto off until their owners and admins turn it on. Lowering it takes effect at the next step of every request.