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

# Needs your input: approve commands, answer questions, provide secrets, and help manually

> The requests that appear when the agent pauses for you: command approval, answering questions, adding secrets, handling a step yourself, sensitive location access requests, confirmation before sending email, and how you find out when you are not in the session.

**Needs your input** is the state in which the agent pauses during a session run and waits for you to decide or provide information, for example to approve a command, answer a question, provide a secret, or complete an operation it cannot do for you, such as signing in. The session continues automatically after you respond.

## Where this applies

| Form | Support | Notes |
| - | - | - |
| Desktop app | Supported | When the session runs on this computer, sensitive location access requests also pop up on this computer |
| Web app | Supported | |
| Mobile web | Supported | |

## Where requests appear

* **Intervention card**: a card directly above the message box on the session page. All requests appear here. When there are several, you page through them in the same card.
* **Timeline**: except for command approvals, a request leaves a line where it occurred, such as **Waiting for your reply** · **Respond**. Click **Respond** to expand the intervention card. After a request is handled, the timeline keeps a record line.
* **Confirmation before sending email (Gmail)** is the exception: its confirmation card shows directly in the timeline. See below.

## Common actions on the intervention card

| Action | How |
| - | - |
| Submit / Allow | The main button at the bottom right of the card, or `⌘Enter` (Windows `Ctrl+Enter`) |
| Deny / Skip / Not now | The secondary button at the bottom of the card |
| Handle later | The × at the top right of the card, or `Esc`. The card collapses into "Needs your input · N pending". Click it to expand again |
| Switch between items | **Previous request** and **Next request** in the card header. You can handle items in any order; the session continues only after all are handled |

* If your reply is not delivered, you see **Reply timed out. You can retry.** If the item has already been handled on another device, you see **Handled on another device**.
* Switching to **Full access** does not automatically approve requests already waiting. You still need to handle them one by one.

### Keyboard

When focus is not in the message box, you can handle the current item directly with the keyboard:

| Request | Keys |
| - | - |
| Command approval | `Enter` or `1` to allow, `2` to allow for this task, `Esc` or `3` to deny |
| Questions and choices | `1`–`9` to choose option N, `Enter` to submit or continue, `Esc` to skip the whole card |

## Request types

| Type | Card title | Main button | Other buttons |
| - | - | - | - |
| Command approval | **Approval** | **Allow** | **Deny**, **Allow for this task**, the **Something else…** field |
| Free-form answer | **Question** | **Submit** | **Skip** |
| Choice | **Choose an option** | **Submit** (**Continue** when there are several questions) | **Skip**, **Previous** |
| Add secret | **Add secret** | **Submit** | **Skip** |
| Needs you to handle it yourself | **Needs your input** | **Done** | **Not now**, **View cloud server**, **Take control** |
| Access request | **Access request** | **Allow once** | **Deny** |

### Command approval

When the permission is **Ask for approval**, or when **Approve for me** detects a possibly unsafe operation, the agent asks for your approval before running it. The card shows the command to run.

* **Allow**: approves only this once.
* **Allow for this task**: approves this once, and similar operations later in this session are allowed automatically.
* **Deny**: the command does not run, and the agent tries another way to continue.
* **Something else…**: write your request and press `Enter`. This denies the command and tells the agent your request.

For choosing permissions, see [New session](https://docs.aihubmix.com/en/cloud/sessions/new-session.md).

### Answer questions and choices

When the agent lacks information, it asks you directly. When it needs you to pick a plan, it gives you choices.

* A card may contain several questions. One shows at a time, and you can click **Previous** to review.
* Choosing does not send automatically: click **Continue** for each question, and click **Submit** on the last one.
* When the question allows it, the options end with **Other…**, where you can write your own answer.
* Questions that require confidentiality use a password field, and the answer does not remain on the page after you submit it.

### Add secret

When the agent is missing a secret (for example, `SENTRY_AUTH_TOKEN`), it asks you to provide it. The card shows the reason, the secret name, and a field, with the note **Used only for this task. Its value won’t appear in the chat.** If you collapse the card or switch to another request, you need to enter it again. For secrets you need long term, save them on the secrets page in settings.

### Needs you to handle it yourself

Some steps the agent cannot complete for you, such as signing in and entering a two-factor code. The card states what you need to do.

1. For sessions on the cloud server: click **View cloud server** or **Take control** to open the cloud server view. For sessions running on a device: operate directly on that computer.
2. Complete the operation. The agent does not pause while you operate the cloud server; see [Cloud server desktop](https://docs.aihubmix.com/en/cloud/cloud-host/desktop.md).
3. Return to the card and click **Done**. The agent continues.

If the step is not needed, click **Not now**, and the agent tries another way to continue.

### Sensitive location access requests

When the session runs on your device and that device's **Sensitive locations** is set to ask every time, the agent asks you before accessing such locations (for example, SSH, the keychain, and system configuration):

* The card shows the reason and the path, with an expiry countdown.
* **Allow once** allows only this access and does not change the device's permission settings. When you allow it on the device performing the access, the system asks for confirmation once more.
* If it expires before you respond, it is handled as denied, and the timeline records **Access request expired; access was not granted**. When needed, send another message asking the agent to retry.
* You can also respond in the web app and on other devices. The desktop app on the device performing the access shows a confirmation at the same time, and the first answer applies.

For a device's **Sensitive locations** setting, see [Run environment](https://docs.aihubmix.com/en/cloud/projects/run-environment.md).

### Confirmation before sending email (Gmail)

After you connect Gmail, the agent always asks you to confirm before sending an email. A **Confirm before sending** card appears in the timeline, listing the sender, recipients, CC, BCC, subject, and a body summary. The email is sent only when you click **Send**. Click **Don’t send** and it is not sent. For connecting Gmail, see [Connectors](https://docs.aihubmix.com/en/cloud/integrations/connectors.md).

## Relationship with the composer and stopping

* While a request is waiting for you, the message box works as usual; see [Composer](https://docs.aihubmix.com/en/cloud/sessions/composer.md).
* While the agent is waiting for you to handle a step yourself, clicking **Steer** on a queued message first ends the request as **Not now**, then inserts the message.
* Clicking **Stop this turn** cancels requests that are still waiting.

## How you find out when you are not in the session

| Location | Shows |
| - | - |
| Sidebar | A **Needs your input** label at the end of the session row; it also shows to the right of a collapsed project's title |
| Shortcut | `⌥⌘A` (Windows `Ctrl+Alt+A`) jumps to the next session waiting for you |
| System notifications | **Approval needed**, **Input needed**, **Action needed**. See [Notifications](https://docs.aihubmix.com/en/cloud/settings/notifications.md) |

Viewing a notification does not handle the request. You still need to return to the session and handle it.

## FAQ

### I collapsed the intervention card. How do I get it back?

Click "Needs your input · N pending" above the message box, or click **Respond** on that line in the timeline.

### What is the difference between Allow and Allow for this task?

**Allow** approves only this once. **Allow for this task** also allows similar operations later in this session automatically. It applies only to the current session.

### If I deny, will the session fail?

Not necessarily. The agent tries to complete the task another way. If it really cannot continue, it explains in its reply. To tell the agent what to do, use **Something else…**.

### Will the secret I enter appear in the chat history?

No. The chat records only "secret name provided. The value stays out of the chat."

## Related

* [New session](https://docs.aihubmix.com/en/cloud/sessions/new-session.md)
* [Composer](https://docs.aihubmix.com/en/cloud/sessions/composer.md)
* [Errors and recovery](https://docs.aihubmix.com/en/cloud/sessions/errors-and-recovery.md)
* [Run environment](https://docs.aihubmix.com/en/cloud/projects/run-environment.md)
* [Cloud server desktop](https://docs.aihubmix.com/en/cloud/cloud-host/desktop.md)
* [Notifications](https://docs.aihubmix.com/en/cloud/settings/notifications.md)


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