Browser Tasks

Browser Tasks let you automate interactions with websites that don't have APIs. Describe what you want to do in plain English, and AI-powered automation handles the clicking, typing, and navigation.

How Browser Tasks Work

1. Start a session — A cloud browser opens in the background
2. Describe the task — Write what you want to accomplish in natural language
3. AI executes — The automation navigates, clicks, types, and extracts data
4. Get results — Data and session state return to your workflow

You don't write selectors or manage browser drivers. The AI interprets your instructions and figures out how to accomplish them.

The Browser Task Action

Add Browser Task to your workflow and configure:

• Task description — What you want the browser to do
• Starting URL — Where to begin (optional if continuing a session)
• Variables — Data from your workflow to use in the task

Example: Log into a Portal and Download a Report

Go to https://vendor-portal.example.com
Log in with username {{trigger.username}} and password {{trigger.password}}
Navigate to the Reports section
Download the monthly sales report for {{trigger.month}}

The AI handles finding the login form, entering credentials, navigating menus, and clicking download buttons.

The Extract from Website Action

Extract from Website is for structured data extraction from web pages. Whereas Browser Task handles general automation (clicks, typing, navigation), Extract from Website navigates to a URL and uses AI to pull out data matching a JSON schema you define.

Add Extract from Website to your workflow and configure:

• Starting URL — The page to visit
• Instruction — What to extract and where to look (e.g. "Extract the product title, price, and availability from the main product section")
• Requested Schema — JSON schema defining the structure of the extracted data
• Selector — Optional CSS selector to scope extraction to a specific element
• Timeout — Optional timeout in milliseconds (default 30000)

Output

• Success — parsed contains the extracted data conforming to your schema, plus optional sessionId for the Browserbase session
• Error — error describes what went wrong, plus optional sessionId for debugging

Example: Extract Product Details

Instruction: Extract the product name, current price, and in-stock status from the product details section
Requested Schema: { "type": "object", "properties": { "name": { "type": "string" }, "price": { "type": "number" }, "inStock": { "type": "boolean" } } }

Use Browser Task when you need multi-step automation (login, navigate, click). Use Extract from Website when you need structured data from a single page.

Use Cases

Session Management

Browser sessions can span multiple actions:

1. First Browser Task logs in and navigates
2. Workflow does other processing
3. Second Browser Task continues in the same session

The session ID passes between actions, maintaining login state and context.

Credentials and Security

For sites requiring login:

• Pass credentials via workflow variables (from triggers or integrations)
• Don't hardcode passwords in task descriptions
• Use secure storage for sensitive data

Browser Tasks run in isolated cloud environments—they don't have access to your local machine.

Best Practices

Limitations

• JavaScript-heavy sites — Most modern sites work, but some complex SPAs may have issues
• CAPTCHAs — Cannot solve CAPTCHAs automatically
• Two-factor auth — Needs workarounds (app passwords, SSO, etc.)
• Site changes — Website updates may require task description updates

Example: Competitive Price Monitoring

1. Scheduled Trigger — Run daily
2. Browser Task — Visit competitor sites, extract current prices
3. Generate Object — Compare to your prices, flag significant changes
4. Condition — If price change > 5%
5. Send Notification — Alert the pricing team