Project openclaw - setup agent with headless browser
🤖 Mastering OpenClaw: Managed Mode & Remote VPS Monitoring
Welcome to this technical deep dive into OpenClaw! If you are setting up an autonomous web agent, choosing the right connection method and monitoring it on a headless server is crucial. This guide covers the differences between agent modes and how to “see” what your agent sees on a remote VPS.
🏎️ 1. Webcast vs. Chromium: Which Mode to Choose?
When configuring OpenClaw, you have two primary ways to let the agent interact with the web:
Method A: Chromium/Chrome (Managed Mode) 🛡️
In this mode, OpenClaw spawns a completely separate browser instance.
- Isolation: It uses its own profile, meaning your personal cookies, history, and passwords stay safe and separate.
- Stability: Controlled via the Chrome DevTools Protocol (CDP), making it faster and more reliable for unattended automation tasks.
- Pro Tip: Using Google Chrome instead of the raw Chromium often yields better results, as Chrome supports more modern web standards and proprietary codecs, helping the agent avoid bot-detection.
Method B: Browser Relay (Extension Mode) 🔌
This method lets the agent “hijack” your existing browser via a Chrome extension.
- Access: It can use tabs you already have open and act on websites where you are already logged in.
📸 2. Visual Debugging: Automated Screenshots
When running “headless” on a VPS, screenshots are your best friend. OpenClaw can be configured to capture the browser’s view at specific intervals or upon specific actions.
How to use it:
- Automatic Captures: By default, OpenClaw often saves screenshots of the current page state to the
workspace/screenshots/orworkspace/memory/folder. - Instruction-based: You can explicitly tell the agent to “take a screenshot” as part of its task. This is useful for verifying that a complex UI has loaded correctly.
- Reviewing: These files are typically saved as
.pngor.jpg. If you are on a VPS, you can download them via SCP or view them directly through the OpenClaw Dashboard session history.
This is the perfect middle ground between “blind” automation and a full live stream, as it provides a historical record of exactly what the agent encountered.
🕵️ 3. How to Monitor Your Agent
Since you’ve opted for Chrome in Managed Mode, you get superior tracking and logging:
- The OpenClaw Dashboard: Accessible via the OpenClaw Interface.
- CDP Logs: Detailed logs of every click and network request can be found in your terminal or under
~/.openclaw/workspace/memory/.
🌐 4. Remote Monitoring: Running “Live” on a Headless VPS
Even without a GUI, you can watch the agent work in real-time using an SSH Tunnel.
Step 1: Create an SSH Tunnel to the Dashboard
On your local computer, run: ssh -L 18789:localhost:18789 user@your-vps-ip Then open http://localhost:18789 locally.
Step 2: Advanced Remote Debugging
- Start Chrome on VPS with:
--remote-debugging-port=9222 - Tunnel the port from your local machine:
ssh -L 9222:localhost:9222 user@your-vps-ip - Inspect locally: Open Chrome and go to
chrome://inspect. Addlocalhost:9222to “Discover network targets” to see the live DevTools stream. 📺
💡 Summary
For the best balance of security and performance, use Chrome in Managed Mode. Leverage automatic screenshots for history and SSH Port Forwarding for live troubleshooting.
Happy Scraping! 🕸️
Last updated: February 2026
Source: Cantaloop Aps.