Browser Workers: Gemini and ChatGPT Web

Zenborg allows you to use your web subscriptions to Google Gemini and OpenAI ChatGPT as automated workers in loops and chats.

This feature enables you to run long plan drafts, syntheses, and code reviews without paying API token rates or needing API access keys.

---

Dedicated Chrome Profile and Isolation

Browser workers do not touch your everyday web browsing session or personal Chrome profile.

• Isolated User Data Directory: Stored by default in ~/.zenborg-browser. Your personal browser extensions, browsing history, and personal cookies are completely separated.
• Chrome Remote Debugging: Zenborg controls this browser instance via Chrome DevTools Protocol (CDP) in a sandboxed session.
• Persistence: Once you log in inside the Zenborg browser window, session cookies and authentication tokens persist across runs.

---

Allowed URL Rules

To keep browser workers focused and secure, navigation is constrained to verified provider paths:

ChatGPT Web Rules

• Origin: https://chatgpt.com
• Allowed Paths: Root conversation (/) or custom GPT endpoints (/g/*).

Gemini Web Rules

• Origin: https://gemini.google.com
• Allowed Paths: App chat (/app and /app/*), custom Gems (/gem/*), and NotebookLM integration (/notebook/*).

---

First-Time Login Walkthrough

Before running a browser worker loop, you must log in once:

1. Open the Loops page or Browser Settings in Zenborg.

2. Click the button labeled “Open browser to log in”.

3. Zenborg launches the dedicated Chrome window.

4. Sign in to your Google account (for Gemini) or your OpenAI account (for ChatGPT).

5. Once your web chat interface loads, close the browser window or leave it running.

6. Zenborg will now connect automatically during loop stages.

---

Handling CAPTCHAs and Login Challenges

Web interfaces occasionally present CAPTCHAs or require periodic re-authentication. Zenborg handles these situations with two distinct mechanisms:

Scenario A: Blocking Login Banner (No Fallback Configured)

If no alternative fallback agent is configured:

1. The loop pauses immediately.
2. A banner appears in the loop view stating: "<Provider> needs you to log in" (e.g., "Google Gemini needs you to log in").
3. The banner displays two buttons:
   • "Open browser to log in": Opens the Chrome profile so you can solve the CAPTCHA or sign in.
   • "Resume loop": Resumes the paused stage once you finish solving the prompt.

Scenario B: Non-Blocking Fallback Notice (Fallback Configured)

If you have configured a fallback worker agent (such as Claude Code or another connected CLI agent):

1. The loop does not pause.
2. The browser worker immediately hands off the task to the secondary agent.
3. Zenborg displays an informative notice banner: "<Provider> hit a CAPTCHA; the loop continued on another agent."
4. The loop finishes without interruption. You can solve the CAPTCHA later at your convenience.

---

Best Practices and Rate Limits

• Avoid Manual Typing During Runs: When a loop stage is actively sending prompts or receiving streaming output, avoid typing in that specific browser window.
• Web Rate Limits: Free and paid web accounts have web rate limits (such as hourly message caps on ChatGPT Plus or Gemini Advanced). If a web provider throttles you, Zenborg triggers its Smart Worker fallback.
• Keep Chrome Updated: Zenborg connects to your installed Google Chrome or Chromium binary. Keep your browser up to date for reliable DevTools communication.

---

Next Steps

• Learn about security scanning in Skills and MCP.
• Follow the hands-on tutorial in Run a Loop on Your Gemini Subscription.