I run OpenClaw on my Mac Mini as a personal assistant for daily automation. Recently I wanted to give it one more ability: talk to ChatGPT through my actual browser, not the API.
Here's what it looks like. I send a message in Telegram, OpenClaw opens ChatGPT in Chrome, sends the prompt, reads the response, and brings it back:
Why Operate ChatGPT This Way
Some ChatGPT models like GPT-5.4 Pro are only available through the web interface with a Plus or Pro subscription. The API has its own model list and own pricing. Going through the browser means OpenClaw gets access to every model I can use, including the ones the API doesn't offer.
When I chat with OpenClaw through Telegram and need research on something, I just ask. OpenClaw opens ChatGPT in the browser, sends my question, reads the response, and replies back to me in Telegram. I don't switch apps or copy-paste anything, and the conversation stays in my ChatGPT history so I can pick it up later.
Because it uses my real browser session, OpenClaw can select any model I have access to. If OpenAI drops a new model next week, OpenClaw just picks it from the dropdown the same way I would.
How I Set It Up: actionbook CLI + Chrome Extension
actionbook is a Browser Action Engine I've been building. It gives AI agents pre-computed "action manuals" for websites, semantic descriptions of what's interactive, so agents don't need to parse raw HTML or guess at selectors.
The key for OpenClaw: actionbook CLI connects to your existing Chrome through a Chrome extension. Not a new browser instance. Your running browser with active session.
There's only one manual step: install the actionbook Chrome extension from the Web Store. Everything else, I just told OpenClaw to do it:
Install actionbook CLI from https://github.com/actionbook/actionbook
run `actionbook setup`, and pick extension mode for browser connection.
During setup, the wizard detects your environment: OS, shell, installed browsers. The important step is Browser Mode, where it picks between:
- isolated: Launch a dedicated browser (clean environment, no setup needed)
- extension: Control your existing Chrome
OpenClaw picked extension as instructed. That's the whole point. I want to control the Chrome that's already logged into ChatGPT, not spin up a fresh one.
How OpenClaw Operates ChatGPT
Once actionbook is set up, I just tell OpenClaw what I need:
Use actionbook to open ChatGPT and ask:
what are the latest trends in AI agents for 2026? Bring me the answer.
I don't tell OpenClaw which button to click or where the input box is. OpenClaw reads the action manual for ChatGPT, takes a snapshot of the page to understand the current layout, and figures out how to operate it on its own. It fills in the prompt, clicks send, waits for the response to finish streaming, and reads the result back to me.
Here's what it runs behind the scenes:
actionbook search "chatgpt"
actionbook get chatgpt.com:/:default
actionbook browser snapshot
actionbook browser fill "your question here" --ref-id e3
actionbook browser click --ref-id e4
actionbook browser wait-idle
actionbook browser text
What Else I Use It For
I also use it for parallel GEO testing. When I need to compare how ChatGPT responds to the same prompt under different contexts, I tell OpenClaw to open multiple tabs and send them all at once.
And this isn't limited to ChatGPT. OpenClaw can operate any page I'm already logged into. That's what makes it a real agent.
Top comments (0)