For the complete documentation index, see llms.txt. This page is also available as Markdown.

Web Agent

CodeWords's cloud-based browser that operates like a real person

Overview

The Web Agent is your AI-powered browser. Instead of writing scripts or clicking through pages manually, you describe what you need and the Web Agent figures out how to get it done by navigating pages, clicking buttons, filling forms, and extracting data.

It runs in a remote browser in the background, so it does not take over your computer. You can watch it work in real time through a live browser view, or let it run and come back to the results.

When to use the web agent

Use the Web Agent for:

  • Complex research tasks where you do not know all the steps upfront.

    • Example: "Find the CEO's email on the company website. Check the About page, Team page, or Contact form."

  • Dynamic websites that are JavaScript-heavy or change often (for example, news sites, dashboards, and social platforms).

    • Example: "Search for product reviews and summarize the top five complaints."

  • Authenticated websites behind logins (for example, LinkedIn profiles, private dashboards, and internal tools).

  • Multi-step interactions like filling out forms, clicking through menus, or scrolling to load content.

For simple public pages, use Firecrawl instead. The Web Agent is built for complex tasks. If you only need to scrape a public webpage, Firecrawl is usually faster and lower cost. Ask Cody, and it will pick the right tool.

How it works

1

Describe your goal

Write clear, step-by-step instructions. The more specific you are, the better the results.

"Go to the company website, find the About or Team page, and get the CEO's name and LinkedIn profile URL."

2

The Web Agent browses for you

It opens a remote browser, navigates to the right pages, interacts with elements, and extracts the data you asked for.

3

Watch live or wait for results

You can watch the browser in real time through the live view panel, or let it run and check the results when it is done.

4

Get your results

The Web Agent returns what it found, including text, structured data, or page content, directly in the chat.

Login and authentication

The Web Agent can access websites that require login. When you first need an authenticated site, Cody opens a live browser session where you can log in securely.

After you log in once, session cookies are saved. That means you do not need to log in again for future automations on the same site, because the Web Agent remembers your authenticated session across runs.

Login limitations

The login step currently supports traditional username and password forms only. "Sign in with Google," single sign-on (SSO), and other OAuth-based login methods are not supported. Make sure the website has a standard login form before asking Cody to log in.

Live browser view

When the Web Agent is running, you can watch it in real time. Cody shows a live browser view in the canvas panel on the right side of your screen. You can see what the agent sees, including pages loading, buttons being clicked, and content being extracted.

When the session finishes, you will see "Session Complete." Return to the chat, and Cody will show your results.

Key limitations

Limitation
Detail

Timeout

Runs have a 10-minute limit. If a task takes longer, it stops.

Login method

Only username/password login is supported. OAuth, SSO, and "Sign in with Google" are not.

Speed

Slower than Firecrawl or the Chrome Extension. Best for tasks that need real browser interaction.

Cost

Higher cost per run than simpler scraping tools, because it uses AI to reason through each step.

Web Agent vs other tools

Tool
Best for
Needs login?
Speed

Firecrawl

Public pages and fast scraping

No

Fast

Web Agent

Complex tasks, dynamic sites, and authenticated pages

Once (saved)

Slower

Chrome Extension

Authenticated pages when you are at your computer

Already logged in

Medium

You do not need to choose the tool yourself. Tell Cody what you need, and it will choose the best approach. If a page is public, it will usually use Firecrawl. If it needs a real browser, it will use the Web Agent.

Example prompts

  • "Go to this company's website and find the CEO's name and email."

  • "Log into my dashboard at [URL] and download the latest report."

  • "Search for reviews of [product] on Google and summarize what people are saying."

  • "Go to this LinkedIn profile and extract their work experience."

  • "Fill out this contact form on [website] with these details."

Last updated

Was this helpful?