The problem

You can see the issue in the browser: the button is too wide, the copy is wrong, the spacing is off, or the component needs a quick behavior change. The slow part is finding the right file, line, and context before your coding agent can make a useful edit.

Layrr sits between your browser and local dev server. It injects a small overlay, maps clicked elements back to source, and gives your coding agent the instruction plus the selected code location.

Install

Install Layrr globally:

npm install -g layrr

Usage

Start your app first:

pnpm dev

Then run Layrr against the dev server port:

layrr --port 3000

Layrr opens a proxied version of your app at http://localhost:4567.

In the browser:

1. Click one or more elements.
2. Describe the change you want.
3. Let the selected coding agent edit the source.
4. Preview or revert Layrr edits from the overlay history.

For a local checkout:

pnpm install
pnpm build
node dist/cli.js --port 3000

Options

layrr --port <number> [project-root] [options]

| Option | Description | | --- | --- | | -p, --port <number> | Local dev server port. Required. | | --proxy-port <number> | Layrr proxy port. Defaults to 4567. | | --agent <name> | AI agent to use: claude, codex, or gemini. | | --gemini-model <model> | Save and use a Gemini model, for example gemini-2.5-flash. | | --configure-gemini | Reconfigure the Gemini model and API key. | | --no-open | Do not open the browser automatically. | | -h, --help | Show help. |

Agents

Layrr supports:

• claude - Claude Code
• codex - Codex CLI
• gemini - Gemini via Pi coding agent, installed as a Layrr dependency

If no agent is configured, Layrr prompts you to pick one.

To configure Gemini without starting a session:

layrr --configure-gemini

Git History

Layrr uses git as its undo path:

• initializes a git repo if needed
• creates an initial snapshot when needed
• commits successful edits with a [layrr] prefix
• keeps pre-existing dirty files out of Layrr edit commits
• lets the overlay preview and revert Layrr edits

License

MIT