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

# Modes and approvals

> Choose how much the AI Assistant does on its own, review its changes, and undo them 🛡️

The AI Assistant can change things in your account, so you decide how it works: whether it only plans, asks before every change, or builds a plan you approved on its own. Whatever the mode, some actions always stay with you.

## Modes

Pick the mode from the **mode picker** in the composer. You can change it at any point in a chat.

| Mode | What the assistant does |
| - | - |
| **Plan** | Researches and asks questions. It changes nothing until you approve its plan and switch to building. |
| **Create** | Builds what you ask, and stops on an **approval card** before every change. |
| **Auto** | Plans with you first — questions and a plan card. Once you approve the plan, it builds it **on its own**, without approval cards, until it's done. |

While Auto is on, the mode picker shows an **Auto** badge. A new message in Auto starts a new task, so the assistant plans it with you again.

New chats start in **Auto** on the Agents and Tools pages and on the full-page assistant, and in **Create** everywhere else.

### Approving a plan

In Plan and Auto, the assistant finishes planning with a **plan card**: the steps it will take, sometimes with a diagram. Below the plan, choose how to build it — **Create** (approve each change) or **Auto** (build on its own) — and approve. Reject the plan with a note to have the assistant rework it.

While it builds, the assistant keeps a **Plan** checklist in the chat that shows how many steps are done.

> [!TIP]
>
> Use Plan or Auto for bigger jobs, like building an agent with several tools or a campaign with a multi-step flow. Use Create when you want to check each change as it happens.

## Approval cards

In Create mode, every change waits on a card in the chat. The card tells you what will happen — for example, the flow steps that will change, the campaign settings, the tool's definition, or the people and teams that will be created.

* **Approve** applies the change.
* **Reject** cancels it. Add a note to tell the assistant what to do instead.
* **Allow for this chat** (on some cards) approves the change and stops asking before that same kind of change for the rest of the chat.

When the assistant prepares several changes in the same step, one **Approve** or **Reject** answers all of them.

Changes that ran without a card — in Auto, or after **Allow for this chat** — are marked in the chat with an **Auto** or **Allowed** chip, so you can always see what happened without asking.

## Seeing and undoing changes

After a change, the chat shows a row for it with a link to open it, such as **Open in Tools** or **Open on canvas**. The page you're on refreshes in place, so you see the result right away. The assistant never navigates on its own.

* **Flow edits** show *"Applied N changes to …"*. Click **Show changes** to compare the flow before and after, and **Undo** to take the edit back. If the flow was edited again since, the row says **Edited since** — undo it from the flow builder instead.
* **Meta Business Agent edits** land in the builder as an unpublished draft, with an **Undo** on the card.
* Other changes are drafts you can edit or discard as usual from their own page.

## What stays with you

The assistant has no way to launch, send or delete anything. These actions always stay a click in the app:

> [card] Campaigns
>
> The assistant creates and edits **draft** campaigns and stops at ready-to-send. You send or schedule them.

> [card] AI agents
>
> Saving an agent creates a **draft version**. Deploying it is up to you.

> [card] Flows and WhatsApp Flows
>
> Edits go to the **draft**. Launching a flow or publishing a WhatsApp Flow is up to you.

> [card] Meta Business Agent
>
> Edits stay an **unpublished draft** in the builder until you click **Publish**.

The assistant also has no tool to delete or archive anything, and in the Inbox it can't send messages, assign, tag or close conversations.

> [!WARNING]
>
> A few changes take effect as soon as you approve them, because these areas have no draft:
>
> * **Knowledge:** a new knowledge source, or a change to which agents read it, changes what your **deployed** agents answer right away. The card says so.
> * **Teams and users:** approving a team setup creates the folders, teams and users at once, and each new user gets an email to set their password.
> * **Tools:** a new tool is created in your account once it passes a live test against your API. Your agents only use it after you add it to them.


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