# Build with your Claude plan

Source: https://layerwise.app/docs/build-your-app/claude

Use the Claude Code on your computer to build in Layerwise with your Claude plan instead of Layerwise credits.

If you use Claude Code, you can build in Layerwise with it. **My Claude** chats run on the Claude Code installed on the computer and draw on its Claude plan instead of the team's Layerwise credits. Any chat can switch between Layerwise, My Codex, and My Claude at any time.

My Claude works on every Layerwise plan.

## Before you start [#before-you-start]

- Install Claude Code on your computer. Anthropic's [setup guide](https://code.claude.com/docs/en/setup) covers macOS, Linux, and Windows.
- Sign in by running `claude auth login` in a terminal. You can also sign in with an Anthropic API key, Amazon Bedrock, Google Vertex AI, or Microsoft Foundry. With those, your provider bills you for usage.

## Check Claude Code [#check-claude-code]

Open **Account** from the user menu and find the **My Claude** card. It shows the signed-in account, its plan, and its usage limits. To see where Claude Code is installed, its version, and how it signs in, open the card's menu and choose **Connection details**.

Layerwise uses Claude Code's own sign-in, so there's nothing to connect. If Claude Code isn't installed, choose **Install** on the card. Once it's installed, the card finds it when you come back to Layerwise. If Claude Code isn't signed in, the card says what to fix. Fix it, then choose the card's refresh button.

Each teammate builds with their own Claude account on their own computer.

## Build with My Claude [#build-with-my-claude]

1. When you create a project or start a chat, open the model picker.
2. Under **Agent**, choose **My Claude**.
3. Choose a **Model** and an **Effort** level. The list comes from Claude Code.

Layerwise remembers your choice for new projects and chats. A My Claude chat works like any other chat, with the same project tools, preview, approval modes, and [queued prompts](/docs/build-your-app/direct-the-agent#queue-your-next-request). You can edit an earlier message, and if you change the model or effort while Claude works, Claude switches right away.

When Claude asks you a question or proposes a plan, the chat shows it and waits for your answer. Goals aren't available in My Claude chats.

## Claude Code setup [#claude-code-setup]

My Claude chats load the Claude Code setup the same way a terminal session does: `CLAUDE.md` files, settings, skills, subagents, plugins, and MCP servers, plus the project's own `.claude` folder. [Layerwise skills and MCP servers](/docs/build-your-app/skills-and-mcp) are there too.

A few things are different in Layerwise:

- A project's `.mcp.json` servers start only if you approved them in Claude Code.
- Claude.ai connectors stay off. Turn them on under **MCP servers** in the **My Claude** card's menu.
- Claude's auto memory is off.

To turn off a skill or MCP server for Layerwise chats, open the **My Claude** card's menu, choose **Skills** or **MCP servers**, and use its switch. The change applies from the chat's next message. To add, edit, or remove one, use Claude Code. Layerwise never edits Claude Code settings or files.

## Approval modes [#approval-modes]

My Claude follows the chat's [approval mode](/docs/build-your-app/approvals-and-changes):

| Mode                 | What Claude does                                                                                                                                    |
| -------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Ask for approval** | Reads the project, and asks before it edits files or runs commands, even ones the Claude Code settings allow.                                       |
| **Approve for me**   | Edits files and runs commands in the project inside an operating system sandbox with no network access. Claude's auto mode reviews everything else. |
| **Full access**      | Edits any file and runs any command without asking.                                                                                                 |

On Windows, **Approve for me** runs commands without an operating system sandbox, and Claude's auto mode reviews each one. Linux needs `bubblewrap` and `socat` for the sandbox; the **My Claude** card shows when they're missing.

## Switch a chat's agent [#switch-a-chats-agent]

A chat can move between Layerwise, My Codex, and My Claude at any time. For example, switch to My Claude when the team's credits run out, or to another agent when a Claude limit is reached.

1. Open the chat's model picker.
2. Under **Agent**, choose the agent you want.

The chat keeps its history, and the new agent picks up where the conversation left off. It starts with that agent's default model, which you can change in the same picker. The first message after a switch takes a little longer while the new agent reads the chat's history.

Switching happens between messages. Wait for any running task to finish, and send or clear any queued messages.

## Usage and limits [#usage-and-limits]

- **Credits** — My Claude chats don't use Layerwise credits and don't appear in Usage. See [Manage billing and usage](/docs/account/billing-and-usage).
- **Claude usage** — With a Claude.ai sign-in, the **My Claude** card shows how much of each usage limit is left and how long until it resets. When a limit is reached, wait for the reset or switch the chat to another agent.
- **Plan limits** — Your Layerwise plan's daily message limit and project limit still apply. On Free, that's 10 messages a day across all chats.

## Privacy [#privacy]

- Claude Code sends model requests from your computer directly to Anthropic or your cloud provider. Your sign-in stays on your computer.
- Claude Code keeps its own session files on your computer, and Layerwise leaves them in place.
- Like any Layerwise chat, a My Claude chat's history is saved with the project and syncs across devices, so project members can read it.
- Claude Code names new projects and chats from the first message.

## Troubleshooting [#troubleshooting]

| Problem                                         | What to do                                                                                                                                       |
| ----------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
| Claude Code isn't installed or found            | Choose **Install** on the **My Claude** card, then come back to Layerwise once it's installed.                                                   |
| Claude Code isn't signed in                     | Run `claude auth login` in a terminal, then check again.                                                                                         |
| The Claude Code version isn't supported         | Run `claude update`, then check again.                                                                                                           |
| The Claude account reached a usage limit        | Check the reset time on the **My Claude** card, or switch the chat to another agent.                                                             |
| The model is unavailable                        | Choose another model.                                                                                                                            |
| The Claude account for this chat changed        | Stop any running task, then choose **Reconnect My Claude** in the chat. The chat's history stays, and the next message uses the current account. |
| **Approve for me** can't use a sandbox on Linux | Install `bubblewrap` and `socat`, then check again.                                                                                              |