Guide
How action approval works
Workpeg AI can do more than answer — it can carry out actions across your Pegs. The control under the message box sets how much of that it should do without checking with you first. There are three modes; Request approval is the narrowest, it is the one you start on, and staying on it is the right answer unless you have a specific reason to widen it.
What action approval is
Workpeg AI is not only a chat window. As it says on its own landing screen, it "can perform actions across all your Pegs" — so at some point the question stops being what should it say and becomes what should it be allowed to do.
Action approval is where you answer that. It is the control in the row underneath the message box, immediately to the right of the + button, and it is labelled with whatever mode is currently in effect — which is why, before you touch it, it reads Request approval. Open it and the dropdown puts the question plainly at the top: How should actions be approved?
One thing the product does not currently define anywhere is what counts as an "action". There is no published list and no examples beyond the sentence quoted above. This page will not invent one for you.
The three modes
The dropdown offers three, listed least permissive first. Each row shows an icon, the mode's name and a single line describing it; those lines are quoted below exactly as the product words them, because the wording is the whole of what you are being told.
Request approval — "Ask before every action"
The narrowest of the three, and the one Workpeg AI starts on. The promise is unqualified: every action, with no carve-out for the ones that look routine. Nothing is meant to happen that you have not seen first.
Stay here if you want the whole picture, if you are new to Workpeg AI, or if the Pegs it can reach hold anything you would not want touched unattended.
Approve for me — "Only ask for actions detected as potentially unsafe"
The middle setting. Routine actions are meant to pass without interrupting you, and the interruption is saved for the ones flagged as potentially unsafe. Quieter than Request approval, narrower than Full access.
Read the description for what it leaves out as much as what it says. It does not name what does the detecting, and it does not say what makes something "potentially unsafe" — so there is no way, from the product, to work out in advance which of your actions would be caught and which would sail past. Choose this one on the strength of wanting fewer prompts, not on the strength of knowing where the line sits.
Full access — "Unrestricted access to the internet and any file on your computer"
The widest of the three. Its description is the one to read slowly rather than skim, because it is not a convenience promise. It does not say "fewer interruptions" or "faster". It says unrestricted access to the internet, and to any file on your computer.
Pick it only if that sentence, taken at face value and in full, describes something you are willing to hand over. If it makes you hesitate, that hesitation is the correct reading of it.
Which one to pick
- Request approval — the default, and the right answer for almost everyone. Pick it if you want to see everything before it happens.
- Approve for me — pick it when being asked every single time is genuinely getting in your way, and you accept that some actions will then go ahead without asking you, on criteria you have not been shown.
- Full access — pick it only if you accept its description as written, unrestricted and unqualified.
There is no "recommended" badge in the product. The dropdown's
heading is followed by a Learn more link, but
that link points at # — there is nothing behind it
today. Clicking it does close the dropdown, though only because
a click anywhere on the page closes it; the link itself is not
intercepted, so the browser follows it like any other and the
address bar picks up a trailing #. Which leaves the
ordering as the only steer you actually get: narrowest at the
top, widest at the bottom.
How to change it
- Find the control in the row of buttons under the message box, just right of +. It is labelled with the mode in effect right now.
- Click it. The dropdown opens under the heading How should actions be approved? — followed by a Learn more link that currently goes nowhere — with the three modes and their descriptions. The one currently in effect carries a check mark.
- Click a mode. The dropdown closes and the control relabels itself on the spot — the label is the setting, so it always tells you where you stand.
Only one composer dropdown is open at a time: opening this one closes the + menu and the Token Consumption panel, and clicking anywhere else on the page closes it again without changing anything.
Where the setting is remembered
Shorter answer than you might expect, and worth knowing before you rely on it:
- It is not per conversation. The mode applies across the Workpeg AI window you have open, and moving between conversations inside it keeps your choice.
- It is not saved to your account. Choosing a mode changes nothing for your colleagues, and nothing on your other devices.
- It is not saved in your browser either. Reload the page, open Workpeg AI in a second tab, or come back tomorrow, and it is back to Request approval.
That last point cuts both ways. Widen the setting and it will not stay widened — but tighten it and that will not stay either. Read the label on the control rather than assuming your last choice survived.
Incognito chats
In an incognito chat the control is not there at all. The reason is written on the new-chat screen, inside the same bordered card as the message box and directly beneath it: This chat will not be saved, and Workpeg AI won't be able to execute actions. That line belongs to the new-chat screen only — open an incognito conversation and it is not repeated there, though the control stays hidden.
With no actions to approve there is nothing for the setting to govern, so it is hidden rather than shown greyed out. Leave the incognito chat and it reappears, on whatever mode was last chosen in this browser tab.
What this page cannot promise
Everything above describes what the three modes say. Here is the part a guide owes you, stated plainly: in the version of Workpeg AI this page describes, choosing a mode records your choice and relabels the control, and we cannot point at anything further that it changes.
- Your choice does not leave the browser tab. It is not attached to the requests Workpeg AI makes when it generates a reply — those carry the message, the model and the kind of conversation, and nothing about approval. So nothing on our side is currently reading your selection and acting on it.
- There is no approval prompt yet. Nothing in Workpeg AI stops to ask you to confirm or refuse a specific action. If a mode's wording leads you to expect you will be asked, nothing that exists today will do the asking.
- "Potentially unsafe" has no published definition. No criteria, no detector named, no way to check your own case against it.
The practical consequence: do not treat Request approval as a barrier that will stop something, and do not read Full access as having granted something. Neither position currently changes what Workpeg AI is able to do. Your real protection today is the same as it was before the control existed — what Workpeg AI is built to reach, and what your workspace has given it.
We would rather say this than let a control that looks like a security boundary be read as one it is not yet enforcing. When the modes start being enforced, this page will be updated to say so, and to say what each one then guarantees.
Questions about what Workpeg AI can reach in your workspace, or about widening or restricting it, are best sent to contact@workpeg.com.