Getting Started

Installation

MoleClaw is currently installed by building from source.

1. Clone the Repository

git clone https://github.com/clark-maybe/mole-extension.git
cd mole-extension

2. Install Dependencies

npm install

3. Build the Extension

npm run build

Build output goes to the build_version/mole-extension_1.0.0/ directory.

4. Load into Chrome

1. Open Chrome and navigate to chrome://extensions/
2. Enable Developer mode in the top right corner
3. Click Load unpacked
4. Select the build_version/mole-extension_1.0.0/ directory
5. The extension is installed — the icon appears in the browser toolbar

First Use

Configure Your LLM

Mole needs to connect to an OpenAI API-compatible LLM service. Complete the configuration before first use:

1. Click the Mole icon in the browser toolbar to open the popup
2. Go to the Options page (right-click the extension icon > Options)
3. Fill in the following settings:
   • API Endpoint — LLM service URL (default: https://api.openai.com/v1)
   • API Key — Your API key
   • Model — Model name to use (e.g., gpt-4o-mini, gpt-4o, etc.)
4. Click Save

TIP

Mole is compatible with any OpenAI API-format service, including but not limited to: OpenAI, Azure OpenAI, Claude (via compatibility layer), locally deployed Ollama, etc.

Summon the AI Assistant

After configuration, visit any webpage:

• Keyboard shortcut: Press Cmd+E (Mac) or Ctrl+E (Windows/Linux) to summon the search box
• Floating ball: A capsule-shaped floating ball appears on the right side of the page — hover to slide out, click to open

Basic Interaction

1. Summon the search box via shortcut or floating ball
2. Type natural language instructions, for example:
   • "Take a screenshot of this page"
   • "What is this page about"
   • "Search for iPhone 16 on Amazon"
   • "Extract the table data from this page"
3. The AI will automatically choose the right tools to execute the task
4. Tool call status is displayed during execution, and results stream back in real time

Task Levels

Mole automatically classifies requests by complexity:

• Direct answer — Simple Q&A, no tools needed
• Single-step — One tool call is sufficient
• Multi-step — Multiple tools working together
• Compound task — Split into independent subtasks executed in parallel