> ## Documentation Index
> Fetch the complete documentation index at: https://docs.workser.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Approvals: what waits for your yes

> Every kind of ask in Workser, the passkey for money, Always allow, trusting a contact, and where the answer goes after you give it.

**An AI agent asks before anything risky and does everything else on its own.** The ask appears above the message box and waits
in **Requests**; answering it lets the Task carry on. This page is the detail behind
[Approvals and Stuck](https://docs.workser.ai/ai-workforce/approvals-and-stuck.md).

## What always asks

* **Money:** refunds, card purchases and other payments from \$50.
* **Deletes and cancellations**, and anything that cannot be undone.
* **A first email or message to someone you have not trusted yet.**
* **Your Never rules**, checked before every action they cover.
* **Any kind of work set to Asks first** in the AI agent's [Autonomy](https://docs.workser.ai/using/autonomy-and-probation.md).

## The four asks

| Head | What it asks | Your answers |
| - | - | - |
| **Question for you** | A choice or a written answer | A chip (its words are the answer), or "Or say it in your own words" |
| **Needs your OK** | What it will do | **Yes, go ahead** / **No**, or **Allow once** / **Don't send** |
| **Needs your permission** | A step on a computer or in an app | **Allow** / **Deny**, and sometimes **Always allow** |
| **Stuck and waiting for you** | What happened, what it tried, what it needs | **Carry on**, or tell it what to do next and **Send** |

An approval shows the action in the AI agent's words: **"If you approve, it will …"**, with **Change → Approve with changes**
when the action can be adjusted.

## Money: the passkey

A purchase or a payment never completes in the chat. **Approve with passkey** opens the card service's own page in a new tab,
with the note **"Approve it with your passkey in the tab that opened. It leaves this list once it's paid."** If the browser
blocks the tab, the card shows a plain **Open your passkey page** link instead.

Only an owner or admin can approve; a member sees **"An owner or admin approves this."**

## Always allow, and trusting a contact

Two answers make future asks rarer, and both are deliberate:

* **Always allow for \<name>** (owner or admin, only when offered): that AI agent may do that kind of step without asking.
  The note says "You can undo it any time." — the control is in its **Autonomy** tab.
* **Approve and trust \<address>** (or **Approve and trust these contacts**): future email to that address goes out without
  asking. "Trusting a contact means their emails go out without asking. You can change this in Logins."

**Trusted contacts** live in **Settings › Identity › Email**. Add an **Address** or a **Whole domain** (shown as `@shop.com`);
a domain matches only that exact domain. Empty: **"No one is trusted yet. Replies wait for your yes."** Removing one asks
**"Stop trusting \<x>?"** and "Their emails will wait for your yes again."

A learned approval can also offer **Let Workser decide these**; the note says Workser decides from then on, and the control is
in that AI agent's **Rules & alerts**.

## Sign-ins and secrets

Some asks carry their own secure form:

* **Save your sign-in for \<site>** asks for a username or email, a password (with Show/Hide), an optional 2FA key, then
  **Save and carry on** or **Skip**. The login goes to the Logins vault and only "Saved, carry on" is returned — never the
  password.
* A secret typed for one Task says **"Sent straight to the task. Workser doesn't keep it."**
* A site that needs a sign-in in **Requests** offers **Sign in** or **Not now** ("Okay. It asks again tomorrow if it still
  needs it.").

## Requests

**Requests** collects every ask across projects, "Most urgent first", with filters **All**, **Approvals**, **Questions**,
**Suggestions**, **Payments** and **Sign-ins**, and **From** for one AI agent.

* Repeated identical approvals read "… asked the same thing N times: “…”" with **Approve all N** or "I'll read them one by
  one".
* Answering one removes it. **"That one is already answered."** means someone else got there first.
* Nothing waiting reads **"Nothing is waiting on you."**

## After you answer

The card collapses in the chat to one quiet line — **Answered** — with the answer word: **You approved**, **You declined**,
**You dismissed it**, **You sent it securely**, **No answer in time**, **Approved automatically**, or **You answered: “…”**.
The AI agent's next message keeps the trail as **"Approved: \<title>"** or **"Declined: \<title>"**, and your reply carries
**"Answer to \<title>"**.

## Cost limits that act like approvals

* **The 10% pause:** a paid Task with no limit of its own pauses and asks after it uses 10% of the month's credits.
* **A Task's own credit limit** (Manage → Advanced) stops it at the limit without asking.
* **Under 10 credits** nothing new starts; the Task stops and **Add credits** appears.
* **The card's monthly limit and stores** can refuse a purchase before the passkey is even offered.

## Telling people it is an AI agent

For chat apps, **Settings › Chat apps → Say it is an AI agent** adds a line to the first message each person gets; some
countries require it. On a call, the legal line is added to the greeting and says the call may be recorded when recording is
on.

## Related

* [Tasks and Requests](https://docs.workser.ai/using/tasks.md): the ask in the flow of a Task
* [Autonomy and probation](https://docs.workser.ai/using/autonomy-and-probation.md): reducing the asks over time
* [Logins](https://docs.workser.ai/using/logins.md): where saved sign-ins and trusted contacts live
* [Approvals and Stuck](https://docs.workser.ai/ai-workforce/approvals-and-stuck.md): the shorter guide


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.