- Nov 18, 2024
- 106
- 57
Hey folks, I've been working on a browser setup for agents (openclaw, hermes etc) that's fundamentally different from all other agent browsers out there. It does not use CDP at all.
It's an interesting and intensive project that I had hacked a prototype together quickly and it worked so well to allow my agents to access social media sites that are normally blocked that I decided to productionize it.
It's a ground up fork of linux with the explicit goal of not using CDP to control the browser, it actually uses google chrome stable, and it does not emulate key presses or anything like that to drive the browser. It's actually a harness around the real browser itself and is hooked in all through the OS level so the actual browser can be fully controlled reliably and deterministically. The agent doesn't know the difference and with some humanization (think random typos, delays) it is working extremely well at just giving access to the agent to use the valuable web.
The reason it's not another anti detect browser is because anti detect browsers are fundamentally flawed, as soon as you fork chromium to comment out navigate.webdriver = true, you've exposed potentially hundreds of signals that this is not chrome, this is a chromium fork. Which most detection systems can see very clearly because google chrome stable is compiled very differently from chromium source code then is available publicly. Some estimates say Chrome now has over 1000 individual signals used for tracking, fingerprinting and bot detection alone.
This is fundamentally different because it is Real google chrome. There's nothing to detect it's the actual browser itself, no different settings, it's the exact package they publish to download Google Chrome. Just driven in a fundamentally lower level way so it does not use CDP and does not trigger webdriver detection. It's otherwise the exact same system and I've setup a streaming setup so you can see agents driving it and what they're seeing, and a cli that is very token efficient catered to agents driving the browser. It's so far working incredibly well with my agents automating various major social media sites without issue and no particular evasion tactics either. Because the browser profile is so strong as a corporate browser, it doesn't even need a residential or cell proxy, I still use a proxy but the activity looks like any other corporate browser.
I'm curious if anyone is interested in this for their agents and would be interested in trying it out very soon? If it's stupid or not needed or you have other solutions that are working well, feel free to give the brutal feedback, I found I didn't want to use other anti detect systems because most of them being a chromium fork were simply not able to evade detection/shadowban, and most other anti detect browsers have a similar problem of the signature is so vastly different it's easily detected as evasive.
Give your agents a real browser is the pitch. Please share your thoughts!
It's an interesting and intensive project that I had hacked a prototype together quickly and it worked so well to allow my agents to access social media sites that are normally blocked that I decided to productionize it.
It's a ground up fork of linux with the explicit goal of not using CDP to control the browser, it actually uses google chrome stable, and it does not emulate key presses or anything like that to drive the browser. It's actually a harness around the real browser itself and is hooked in all through the OS level so the actual browser can be fully controlled reliably and deterministically. The agent doesn't know the difference and with some humanization (think random typos, delays) it is working extremely well at just giving access to the agent to use the valuable web.
The reason it's not another anti detect browser is because anti detect browsers are fundamentally flawed, as soon as you fork chromium to comment out navigate.webdriver = true, you've exposed potentially hundreds of signals that this is not chrome, this is a chromium fork. Which most detection systems can see very clearly because google chrome stable is compiled very differently from chromium source code then is available publicly. Some estimates say Chrome now has over 1000 individual signals used for tracking, fingerprinting and bot detection alone.
This is fundamentally different because it is Real google chrome. There's nothing to detect it's the actual browser itself, no different settings, it's the exact package they publish to download Google Chrome. Just driven in a fundamentally lower level way so it does not use CDP and does not trigger webdriver detection. It's otherwise the exact same system and I've setup a streaming setup so you can see agents driving it and what they're seeing, and a cli that is very token efficient catered to agents driving the browser. It's so far working incredibly well with my agents automating various major social media sites without issue and no particular evasion tactics either. Because the browser profile is so strong as a corporate browser, it doesn't even need a residential or cell proxy, I still use a proxy but the activity looks like any other corporate browser.
I'm curious if anyone is interested in this for their agents and would be interested in trying it out very soon? If it's stupid or not needed or you have other solutions that are working well, feel free to give the brutal feedback, I found I didn't want to use other anti detect systems because most of them being a chromium fork were simply not able to evade detection/shadowban, and most other anti detect browsers have a similar problem of the signature is so vastly different it's easily detected as evasive.
Give your agents a real browser is the pitch. Please share your thoughts!
