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

# Wonder Chat

> Chat with Claude Code or Codex beside your canvas and your code in the Wonder desktop app: start a chat, steer or queue follow-ups, answer the agent's questions and get notified when it finishes.

Wonder Chat is where you work with an agent in the Wonder desktop app. Start a chat, describe what you want, and follow the agent as it designs on your canvas and changes your code.

**Before you start** · Desktop app · Free plan

## What Wonder Chat is

Each chat runs Claude Code or Codex on your own computer, signed in with your own account. Wonder connects it to your canvas through [Wonder MCP](https://wonder.design/docs/wonder-mcp.md) and to your repository, so the agent sees your designs and your code in one place, with the MCP servers and skills you've already set up for it.

Chats run on your own Claude Code or Codex plan, so they use no Wonder credits. On the Free plan, the layers your agents create on the canvas count toward 3,000 AI-generated layers a week. See [Check your AI-generated layers](https://wonder.design/docs/design-with-ai.md#check-your-ai-generated-layers).

To get started, [connect Claude Code or Codex](https://wonder.design/docs/connect-claude-code-or-codex.md).

## Start a chat

* **In a repository**: click **+** (**New worktree**) on the repository's row in the sidebar. The chat works in a new worktree, a separate copy of your code, so several chats can work at once without getting in each other's way.
* **Without a repository**: click **+** (**New quick chat**) next to **Quick chats** in the sidebar, or **Create quick chat** when you have none yet. A quick chat is for asking questions or designing on the canvas when there's no code involved. To add code to it later, choose **Turn into a repository** from its menu.
* **Beside the chat you have open**: click **Create new chat** at the top of the sidebar. It starts a quick chat beside a quick chat, or a chat in a new worktree beside a repository's chat.
* **From the keyboard**: in the chat, press <kbd>⌘</kbd> <kbd>T</kbd> (<kbd>Ctrl</kbd> <kbd>T</kbd> on Windows). It starts a chat in the worktree you have open, or a quick chat if none is. **New chat** is also in the command palette.

A new chat in a repository opens on **What are we making today?**, and a quick chat on **What would you like to work on?**.

Pick the agent and its model from the model picker under the message box. Picking the other agent's model in a chat you've already used opens a new chat beside it. See [Choose models and reviewers](https://wonder.design/docs/choose-models-and-reviewers.md).

Show or hide the chat beside the canvas with <kbd>⌘</kbd> <kbd>⇧</kbd> <kbd>B</kbd> (<kbd>Ctrl</kbd> <kbd>Shift</kbd> <kbd>B</kbd> on Windows).

## Write and send a message

1. Type in the message box. In a repository it reads **Describe what you want to do…**, and in a quick chat **Ask anything or describe a design. No repository attached.**
2. Press <kbd>Enter</kbd> to send. Press <kbd>Shift</kbd> <kbd>Enter</kbd> for a new line.

Press <kbd>⌘</kbd> <kbd>L</kbd> (<kbd>Ctrl</kbd> <kbd>L</kbd> on Windows) to jump to the message box. With layers selected on the canvas, it adds them to the chat instead. Press <kbd>↑</kbd> in an empty box to bring back an earlier message.

To give the agent more to work with, click **+** in the message box or type **@**, and attach files and images, a Figma design or your canvas selection. You can also drop files on the chat. See [Give your agent context](https://wonder.design/docs/give-your-agent-context.md).

## Steer, queue or stop the agent

You can keep typing while the agent works. A message you send then is a follow-up, and it goes one of two ways:

* **Queue** waits until the agent is done, then sends. This is the default.
* **Steer** joins the current work: the agent reads it right away and adjusts what it's doing.

<kbd>Enter</kbd> sends your default way, and <kbd>⌘</kbd> <kbd>Enter</kbd> (<kbd>Ctrl</kbd> <kbd>Enter</kbd> on Windows) sends the other way. Hover over the send button to see which one it does. To change the default, see [Choose what a message sent mid-task does](https://wonder.design/docs/agent-and-chat-preferences.md#choose-what-a-message-sent-mid-task-does). If the agent can't take a message mid-turn, a steered message is queued instead.

Queued messages wait above the message box, under **1 queued** (or how many there are). For each one, you can:

* Click **Steer** to send it into the current work now, or **Send now** once the agent has stopped.
* Click **Edit** to change it, then **Save changes**. Press <kbd>Esc</kbd> to cancel the edit.
* Drag it to reorder the queue, or click **Delete** to remove it.

Click **Clear queue** to remove them all.

![A chat in a repository: the agent is Wondering for 17s, its Todos card shows 0 of 3 done, the last line reads Running a command, and above the message box 1 queued holds the follow-up Also add a one-line note about the license](https://mintcdn.com/wonder-0722b0f6/Ig4tBDjfjeU3SKs6/images/wonder-chat/steer-and-queue.light.webp?fit=max\&auto=format\&n=Ig4tBDjfjeU3SKs6\&q=85\&s=5c81a343affaa3707a854dce794c1633)

![A chat in a repository: the agent is Wondering for 17s, its Todos card shows 0 of 3 done, the last line reads Running a command, and above the message box 1 queued holds the follow-up Also add a one-line note about the license](https://mintcdn.com/wonder-0722b0f6/Ig4tBDjfjeU3SKs6/images/wonder-chat/steer-and-queue.dark.webp?fit=max\&auto=format\&n=Ig4tBDjfjeU3SKs6\&q=85\&s=72393e7b5a0ef4f2aa7201b9084fc2c2)

To stop the agent, click **Stop the agent** where the send button was, or press <kbd>Esc</kbd> in the message box. If messages are queued, they pause with **Paused after you stopped the agent**. Click **Resume** to send them. In a Claude Code chat with nothing queued, **Resume** shows under your last message instead: click it to let the agent carry on where it stopped.

## Follow what the agent is doing

While the agent works, the chat shows how long it's been going, such as **Wondering for 1m 20s**, and its last line says what the agent is doing, such as **Thinking**, **Editing the canvas**, **Editing files** or **Running a command**. When it needs you, it reads **Waiting for approval** or **Waiting for your answer**.

In the chat itself:

* The agent's work between replies is grouped and collapsed, such as **Ran 3 commands, edited 2 files**. Click it to see the agent's thinking and each tool it used, then expand a tool to see what it did. To open them unfolded, see [Expand agent work by default](https://wonder.design/docs/agent-and-chat-preferences.md#expand-agent-work-by-default).
* A **Todos** card shows the agent's plan, with each step **Not started**, **In progress** or **Done**.
* When the agent hands work to a subagent, click **Open conversation** to follow it.
* When the agent designs, a **Designs** card shows how many are ready. See [Watch the agent work](https://wonder.design/docs/design-with-ai.md#watch-the-agent-work).

To search the chat, click in it and press <kbd>⌘</kbd> <kbd>F</kbd> (<kbd>Ctrl</kbd> <kbd>F</kbd> on Windows). **Find in chat** also searches the agent's thinking and the tools it used.

Hover over one of your messages to copy it, go back to before it, or continue from it in a new chat. See [Manage your chats](https://wonder.design/docs/manage-your-chats.md#go-back-to-before-a-message).

## Answer the agent's questions

When the agent needs a decision from you, it asks in a card above the message box.

1. Pick an option, or several when the question allows it. To answer in your own words, type in **Something else…**.
2. Click **Continue**. With several questions, use **Previous question** and **Next question** to move between them.

![A question card above the message box asks whether the README's Getting started section should be brief or detailed, with the options Brief and Detailed, a Something else… field, and the Skip and Continue buttons](https://mintcdn.com/wonder-0722b0f6/Ig4tBDjfjeU3SKs6/images/wonder-chat/answer-a-question.light.webp?fit=max\&auto=format\&n=Ig4tBDjfjeU3SKs6\&q=85\&s=dc6bf5e954a427c697d20f91c7ed88a6)

![A question card above the message box asks whether the README's Getting started section should be brief or detailed, with the options Brief and Detailed, a Something else… field, and the Skip and Continue buttons](https://mintcdn.com/wonder-0722b0f6/Ig4tBDjfjeU3SKs6/images/wonder-chat/answer-a-question.dark.webp?fit=max\&auto=format\&n=Ig4tBDjfjeU3SKs6\&q=85\&s=e5ccca6c7e0d77b3064ca0dc649b77b0)

Click **Skip** to move on without answering. When the agent marked an answer as recommended, open the menu next to **Skip** and choose **Use recommended answers**.

Before some actions, such as running a command, the agent asks for your approval in the same place. See [Approve or deny an action](https://wonder.design/docs/permission-modes-and-approvals.md#approve-or-deny-an-action).

## Slash commands

Type <kbd>/</kbd> at the start of a message to open the list of commands, and keep typing to filter it. Alongside your agent's own commands and skills, Wonder adds these:

| Command | What it does |
| - | - |
| `/add-dir` | Adds a folder to the chat as a working directory, so the agent can read and change it too. |
| `/add-file` | Adds a Wonder file to the worktree, so the agent works in it. Paste a file's link after the command, or leave it out to pick one. |
| `/resume` | Picks up a Claude Code or Codex conversation you started in your terminal. See [Pick up a conversation from the terminal](https://wonder.design/docs/manage-your-chats.md#pick-up-a-conversation-from-the-terminal). |
| `/effort` | Sets how long the agent thinks, for example `/effort high`. The levels depend on the model. |
| `/login` | Signs the chat's Claude Code account in again. Claude Code only. |

In a Claude Code chat, the list also has commands that open in a terminal tab beside the chat, such as `/mcp`, `/permissions` and `/status`.

## Get notified when a chat finishes

When a chat finishes, or needs your approval or an answer, while you're looking at something else, Wonder lets you know:

* A desktop notification names the chat and says what happened: **Finished responding**, **Needs your approval**, **Has a question for you** or **A turn failed**.
* The Wonder icon in the Dock shows how many chats need you. On Windows, the taskbar icon shows a dot.
* In the sidebar, the chat shows a dot until you open it. Choose **Mark as read** or **Mark as unread** from a worktree's menu.

To pick the sound or turn notifications off, see [Get notified when a chat finishes](https://wonder.design/docs/agent-and-chat-preferences.md#get-notified-when-a-chat-finishes) in Agent and chat preferences.

## Chats in the browser

Chats run Claude Code or Codex on your computer, so they're only in the desktop app. In Wonder in the browser, the chats page says **Agent chats run in the Wonder desktop app** and offers the download. Files you open in the browser have no chat.

You can still work on your files from the browser and from other agents over [Wonder MCP](https://wonder.design/docs/wonder-mcp.md), and see the changes chats make, since they work on the same files.

## If something goes wrong

* **The model picker says Not connected.** No agent is signed in. See [Connect Claude Code or Codex](https://wonder.design/docs/connect-claude-code-or-codex.md).
* **The chat says Claude Code isn't signed in, or Codex isn't signed in.** Click **Sign in to Claude** or **Sign in to Codex** in the chat. See [Fix a stopped chat](https://wonder.design/docs/fix-a-stopped-chat.md#claude-code-isn’t-signed-in).
* **Your queued messages don't send.** They're paused, for example after you stopped the agent or a turn failed. Click **Resume**.
* **Send now isn't available.** The reason shows when you hover over it, such as **Answer the agent first**.

## Related

* [Connect Claude Code or Codex](https://wonder.design/docs/connect-claude-code-or-codex.md)
* [Give your agent context](https://wonder.design/docs/give-your-agent-context.md)
* [Manage your chats](https://wonder.design/docs/manage-your-chats.md)
* [Fix a stopped chat](https://wonder.design/docs/fix-a-stopped-chat.md)


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