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

# Approval Mode

> Choose Manual or Auto for the workspace, each run type, one Automation, or one chat, and see what each choice does

The Approval mode decides each agent write that no [Rule](https://docs.cloudthinker.io/guide/approvals/rules.md) covers. Every run uses one of two modes, **Manual** or **Auto**: reads always run in both, and a run no person can answer refuses a write instead of hanging.

## The two modes

| Mode | What happens to a write that no Rule decides |
| - | - |
| **Manual** | A person approves every write before it runs, also one that an **Allow** Rule covers. **Ask first** and **Deny** Rules still apply. |
| **Auto** | The write goes through three steps, in order, until one gives an answer |

Under **Auto**, each write goes through these steps:

1. **Your Rules** allow, ask, or deny the write. This step is always on.
2. **AI judgment** allows a safe write that no Rule covers. You can turn this step off.
3. **If still not allowed**, the write waits for a person (**Ask me**) or is refused (**Refuse**).

> [!NOTE]
>
> Rules decide only when Rule enforcement is on for your workspace. It's off by default, and CloudThinker turns it on per workspace. If the **Rules** step on **Approvals** says **In preview**, Rules only record what they would decide, and the Approval mode alone decides each write. See [Rules in preview](https://docs.cloudthinker.io/guide/approvals/rules.md#troubleshooting).

That gives four Auto choices. The picker names each one by its steps:

| Choice | AI judgment | If still not allowed | Good for |
| - | - | - | - |
| **Auto · Rules · Ask me** | Off | Asks a person | Teams that want only accepted Rules to run writes alone |
| **Auto · Rules · Refuse** | Off | Refused | Unattended runs with a tight, known set of writes |
| **Auto · Rules + AI · Ask me** | On | Asks a person | Most teams. This is the default for new workspaces. |
| **Auto · Rules + AI · Refuse** | On | Refused | Unattended runs. This is the built-in default for Automations. |

AI judgment allows a write that matches what you asked for and stays in its scope. A destructive write, or one with an unclear effect, goes to step 3. If AI judgment cannot finish its check, the write also goes to step 3.

## Set the workspace mode

**1. Open the Approval mode tab**

Open **Approvals** in the sidebar, then the **Approval mode** tab.

**2. Pick the mode**

Under **Workspace mode**, open the picker. Select **Manual**, or select **Auto** and set the **AI judgment** switch and **If still not allowed**.

**Success state:** the picker shows your choice, such as **Auto · Rules + AI · Ask me**. It applies at once, with no confirm step.

You need the workspace Admin role to change a mode. Other roles see "View only. Ask a workspace admin to change these."

## Set a mode for a run type

A run type is where a run starts. Each one can use its own mode, also a looser one than the workspace.

**1. Find the run type**

On the **Approval mode** tab, go to **Run types**. Each row reads **Same as workspace** until you change it, except **Automations**, which shows its built-in **Auto · Rules + AI · Refuse** with a **Built-in** badge.

**2. Override it**

Click **Override** on the row (or the mode shown on **Automations**) and pick a mode. Click **Reset** later to follow the workspace again.

**Success state:** the row shows the mode you picked instead of **Same as workspace**.

| Run type | Default | Can ask a person |
| - | - | - |
| **Chat** | Same as workspace | Yes |
| **Resolve** | Same as workspace | Yes |
| **Optimize** | Same as workspace | No |
| **Automations** | **Auto · Rules + AI · Refuse** (built in) | Yes |
| **Review** | Same as workspace | No |

No person can answer an Optimize or Review run, so **Ask me** becomes **Refuse** there. A Resolve ask, such as one from a webhook incident overnight, waits on its card until someone answers. Turn on **Agent approvals** in [Notifications](https://docs.cloudthinker.io/guide/notifications.md) so someone sees it.

![Approval mode tab with the Workspace mode picker and a Run types list where each row shows Same as workspace or an override](https://mintcdn.com/cloudthinker/7N5Axt6gSxkmeWiW/images/v2/approvals-mode.webp?fit=max\&auto=format\&n=7N5Axt6gSxkmeWiW\&q=85\&s=4c8766185e3ec014623a860638e45501)

Approval mode: the workspace mode plus per-run-type overrides for Chat, Resolve, Optimize, Automations and Review

## Set a mode for one Automation or one chat

* **One Automation:** open the [Automation](https://docs.cloudthinker.io/guide/automation/automations.md) and use its approval picker. **Default** follows the **Automations** run type.
* **One chat:** open the approval control in the chat prompt box, or type `/` and pick **Approval**. **Reset to default** follows the run type and workspace again.

Anyone who can chat in a conversation can change its mode. A chat pick never changes the workspace mode. Before the first message your pick is a draft, and the first message saves it. Each change posts a note in the chat, and the agent follows the new mode from its next step.

Under **Chat composer**, an admin can turn off **Let people pick AI judgment for one chat**. People can then pick only **Manual**, or **Auto** with AI judgment off, for a single chat.

## Which mode a run uses

The most specific mode that someone set wins, even when it is looser:

![The chat pick wins over the Automation's mode, which wins over the run type's mode, which wins over the workspace mode.](https://mintcdn.com/cloudthinker/IOnNsZPtwT4C7tHY/images/approvals/diagram-mode-precedence-light.svg?fit=max\&auto=format\&n=IOnNsZPtwT4C7tHY\&q=85\&s=bd3d730904d50fe73ae6daaff7eb0e56)

The most specific mode that someone set wins, even when it is looser.

1. The mode picked for this chat
2. The mode of the Automation that started the run
3. The mode of the run type
4. The workspace mode

For example, a workspace in **Manual** with **Automations** on **Auto · Rules + AI · Refuse** runs a scheduled Automation in Auto.

> [!WARNING]
>
> A chat pick can loosen a **Manual** workspace, for example to **Auto · Rules · Ask me**. The only admin control over chat picks is **Let people pick AI judgment for one chat** under **Chat composer**. Turn that off and every mode a chat can still pick either asks a person or refuses the write — today that is what keeps a write from running without someone. Once Rule enforcement is on for your workspace, an **Ask first** or **Deny** Rule does the same in every mode (see the note above).

To let an Automation run a routine write every night, accept an **Allow** Rule for that exact operation and connection, scoped to **Automations** under **Runs from**. Then, with Rule enforcement on, the run never depends on AI judgment or on someone being online.

## FAQ

> [accordion] Does Manual turn my Rules off?
>
> No. **Ask first** and **Deny** Rules still apply. Manual only stops an **Allow** Rule or AI judgment from running a write without a person.

> [accordion] Why was a scheduled Automation refused instead of asking?
>
> Automations use **Auto · Rules + AI · Refuse** by default, because no one may be there to answer. Accept a Rule for that write, or set the Automation to an **Ask me** choice.

> [accordion] Why does Ask me show as disabled for Optimize or Review?
>
> No person answers those runs, so a write that is still not allowed is refused. Use a Rule to let a specific write run.

> [accordion] I can't pick AI judgment in a chat
>
> You see "An admin turned off AI judgment in conversations." Ask a workspace Admin to turn on **Let people pick AI judgment for one chat** under **Chat composer**.

> [accordion] A destructive write is waiting for me. Is anyone else told?
>
> Yes. A destructive write that waits for a person also raises a High alert. It reaches every channel on the **Alerts** row in [Notifications](https://docs.cloudthinker.io/guide/notifications.md) whose severity threshold is High or lower.

## Next steps

> [card] [Approval Rules](https://docs.cloudthinker.io/guide/approvals/rules.md)
>
> Accept Rules so common writes do not depend on the mode

> [card] [Approvers and Activity](https://docs.cloudthinker.io/guide/approvals/approvers.md)
>
> Choose who answers each ask and audit what was decided

> [card] [Automations](https://docs.cloudthinker.io/guide/automation/automations.md)
>
> Run agent instructions on schedules, webhooks, and repository events

> [card] [Approvals](https://docs.cloudthinker.io/guide/approvals/overview.md)
>
> See how Rules, the mode, and approvers decide each write


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