Quick Start

This walkthrough uses a normal webpage with a visible search field. The same flow applies to any page the content script can scan.

1. Configure a provider

Open Navvy Settings and create or select a provider profile with:

• baseURL
• model
• optional apiKey

The extension sends the model request to POST {baseURL}/chat/completions.

2. Open a page

Open a product page, documentation page, or search page. Wait for the page to finish rendering before starting the task.

3. Enter a task

Open the Navvy side panel. The composer placeholder is:

Describe your task... (Enter to send)

Try a small task first:

Find the search box and search for wireless headphones.

4. Watch the activity stream

The status dot moves from Ready to Running. During execution, the activity area can show messages such as:

Thinking...
Executing click_element_by_index...
Done: click_element_by_index

Each step records the action name, JSON input, output, and model usage when available.

Expected behavior

Navvy scans the DOM, serializes visible interactive elements with numeric indexes, asks the model for the next AgentOutput tool call, resolves the selected index against the current selectorMap, and performs the action in the content script.

After the page changes, Navvy rescans before planning the next step.

When Navvy asks a question

If the model cannot choose safely, it can call ask_user with a question. That question appears in the UI through the side panel’s ask-user callback, and the run continues after the user responds.