$cd ../use-cases/
πŸ‘Ύ Developer ToolsNew15 min setup
$ cat browser-automation-business.md

await automeBrowser(chrome, { realSession: true })

/** Real Chrome browser control. Navigate, click, extract, fill forms β€” OpenClaw drives the browser like a human. */

how_it_works.md

βš™οΈ How It Works

1.
OpenClaw launches a Chrome session
Since v2026.3.13, OpenClaw can control a real Chrome browser β€” not headless, not Puppeteer scripts. Actual browser sessions with full JavaScript execution.
2.
You describe the task in natural language
"Go to my bank dashboard, download this month's statement, and summarize my spending." OpenClaw plans multi-step navigation automatically.
3.
AI navigates like a human
Clicks buttons, fills forms, handles popups, waits for page loads, scrolls infinite feeds. It can even solve simple CAPTCHAs with vision models.
4.
Extracts and processes data
Scraped data goes directly into your memory, files, or downstream pipeline. Structured JSON output for tables and lists.
IDENTITY.md

πŸ“ Configuration

# IDENTITY.md for Browser Automation Agent

You are a web automation specialist. You control a real 
Chrome browser to accomplish tasks the user describes.

## Browser Rules
1. Always wait for page load before interacting
2. Take screenshots at key steps for audit trail
3. Never auto-submit payment or legal forms
4. Rate-limit requests: max 1 page/3 seconds
5. Respect robots.txt unless user explicitly overrides

## Data Extraction
- Output structured JSON for tabular data
- Include source URL and timestamp with every extraction
- Store in memory for follow-up queries
- For recurring tasks, save extraction templates

## Authentication
- Use credentials from secrets manager only
- Never type passwords visible in screenshots
- Clear session cookies after sensitive operations
- Support 2FA prompts: pause and ask user for code

## Error Handling
- If CAPTCHA detected: attempt vision solve Γ— 3, then ask user
- If login fails: do NOT retry (lockout risk)
- If page structure changed: report and suggest updated selectors
- If rate-limited (429): exponential backoff, max 5 retries
real_world_apps.md

πŸ“‹ Real-World Applications

Price Monitoring

Track competitor prices across 50+ product pages daily. Get alerts when prices drop below thresholds. Export price history to spreadsheet.

Lead Generation

Scrape LinkedIn profiles, company directories, and job boards. Export structured data (name, title, company, email) to your CRM.

Automated E2E Testing

Write tests in plain English: 'Log in, add item to cart, checkout with test card, verify confirmation.' OpenClaw reports what broke.

Government Form Filling

Auto-fill repetitive government or tax forms using your stored data. Screenshot every step for compliance. Review before submission.

Social Media Monitoring

Monitor brand mentions, competitor activity, and trending topics across platforms β€” no API required.

example_session.md

πŸ’¬ Example Session

// User via Telegram:
"Check Amazon prices for RTX 5090 GPUs and tell me if any are under $2000"
// OpenClaw:
β†’ Launching Chrome session...
β†’ Navigating to amazon.com, searching "RTX 5090"...
β†’ Found 23 listings. 3 are under $2000:
1. ASUS TUF RTX 5090 β€” $1,899 (In Stock)
2. MSI Ventus RTX 5090 β€” $1,949 (Low Stock)
3. Gigabyte Eagle RTX 5090 β€” $1,979 (Pre-order)
β†’ Saved results to memory. Want me to set up daily monitoring?
price_monitor.log

πŸ”„ Detailed Workflow: Price Monitoring

Task: Monitor RTX 5090 prices across 5 retailers

Schedule: Daily at 8:00 AM

Run Log β€” March 17, 2026:

[08:00:01] Starting Chrome session (sandboxed)
[08:00:03] β†’ amazon.com/s?q=RTX+5090
  Extracted 23 listings, 3 under $2000
  Screenshot: screenshots/amazon_20260317.png

[08:00:09] β†’ newegg.com/p/pl?d=RTX+5090
  Extracted 18 listings, 1 under $2000
  Screenshot: screenshots/newegg_20260317.png

[08:00:14] β†’ bestbuy.com/searchpage.jsp?st=RTX+5090
  Extracted 8 listings, 0 under $2000
  Note: All models $2,099+ at Best Buy

[08:00:19] β†’ microcenter.com
  Login required β†’ using saved credentials
  2FA prompt β†’ user provided code via Telegram
  Extracted 12 listings, 2 under $2000

[08:00:28] β†’ bhphotovideo.com
  Extracted 6 listings, 1 under $2000  

Summary generated:
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ Retailer    β”‚ Listings β”‚ Under $2K  β”‚ Best   β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚ Amazon      β”‚ 23       β”‚ 3          β”‚ $1,849 β”‚
β”‚ Newegg      β”‚ 18       β”‚ 1          β”‚ $1,929 β”‚
β”‚ Best Buy    β”‚ 8        β”‚ 0          β”‚ $2,099 β”‚
β”‚ Micro Centerβ”‚ 12       β”‚ 2          β”‚ $1,879 β”‚
β”‚ B&H Photo   β”‚ 6        β”‚ 1          β”‚ $1,969 β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Alert sent: "7 RTX 5090 GPUs found under $2,000. 
Best deal: Amazon ASUS TUF at $1,849."

Price trend vs yesterday: ↓ 2.3% average

[08:00:32] Session closed. Next run: tomorrow 8:00 AM

πŸ“Š OpenClaw vs Traditional Tools

FeaturePuppeteer/PlaywrightSeleniumOpenClaw Browser
SetupWrite codeWrite code + driversNatural language
MaintenanceBreak on DOM changesBreak on DOM changesSelf-healing (AI)
CAPTCHANoNoVision model (basic)
Learning curveJavaScript/PythonJava/PythonEnglish/δΈ­ζ–‡/ζ—₯本θͺž
Auth handlingManual codeManual codeSecrets manager
Error recoveryTry-catchTry-catchAI retry + fallback
CostFreeFreeFree (runs locally)

πŸ“ˆ Results: E-Commerce Price Monitoring

MetricManualWith OpenClawChange
Products tracked10200+↑ 20x
Time per check45 min30 sec (auto)↓ 99%
Price alerts1/week (manual)Real-timeInstant
Money saved (deals caught)$0$340/mo avgNew value
Script maintenance2 hrs/week0↓ 100%

πŸ”’ Security Architecture

Browser automation runs in a sandboxed gVisor container. Each session gets isolated: separate filesystem, network namespace, and cookie jar. Credentials are injected from the secrets manager at runtime and never persisted in browser storage. All sessions are recorded for audit (video + action log).

βœ“ Sandboxed gVisor container per session
βœ“ Credentials injected, never persisted
βœ“ Session recordings for audit compliance
βœ“ Rate limiting prevents IP bans
βœ“ Automatic cookie cleanup after sensitive ops

❓ FAQ

Q1. Can it handle JavaScript-heavy SPAs?

Yes. OpenClaw uses a real Chrome browser with full JavaScript execution. React, Vue, Angular, and Next.js apps all work. It waits for dynamic content to render before extracting data.

Q2. Will websites detect and block it?

Unlikely for normal use. OpenClaw uses real Chrome (not headless), with human-like timing and mouse movements. Rate limiting (1 page/3s default) avoids triggering bot detection. For high-volume scraping, you can add residential proxies.

Q3. How does it handle login-required pages?

Credentials are stored in the OpenClaw secrets manager (encrypted at rest). The agent fills login forms automatically. For 2FA, it pauses and asks you for the code via Telegram.