We don't do the forked Chromium patch approach that involves cat and mouse with antibot updates.
We run hardware-backed browsers that are inherently human-like: real consumer components, isp internet, real operating systems. It's much less of a fight for us and that's the value prop for big data at scale, some can't afford that disconnect risk if antibot updates.
Driver is expanding its datacenter footprint 800% this month.
New ISP networks. Racked Android phones. More GPUs.
Daily posts through the whole build. What do you want to see?
@mocokoco2 We run a web infra company, these servers run browser infra, bunch of chromes you can connect to via CDP.
It is for any web automation, scraping included. We specialize in difficult antibot sites (Temu, Shein, Gov lookups, etc).
We also do custom web scraping for people.
A couple options.
The default is unique IP per browser session, location defined by parameter set on launch.
We also have BYO proxies where those can be defined by user. We can also do IPs that are static forever since we own our network (they are residential too!), must contact.
Driver is expanding its datacenter footprint 800% this month.
New ISP networks. Racked Android phones. More GPUs.
Daily posts through the whole build. What do you want to see?
@Stark_Dynamics Browser infrastructure. Real Chrome on real hardware, so you can automate the web without getting blocked. Scraping, agents, anything that needs a browser.
Driver is expanding its datacenter footprint 800% this month.
New ISP networks. Racked Android phones. More GPUs.
Daily posts through the whole build. What do you want to see?
@mkemka_ Actually mobile browsers and future proofing of Play Integrity (key attestation) on the web
Might be used for cloud phone automation too in the future
You can understand a bit more why if you look at driver.dev
We heard browser runtime performance is the new moat?
Driver scores a mean of 15.2 on Browserbench Speedometer. Our fastest hardware tier averages 21.3.
How:
- Dedicated server colocation
- Custom ISO deployed on metal
- Gaming CPUs boosting > 5GHz
Turns out real hardware is fast.
Grok 4.5 has been a delightful surprise.
By far, the most annoying part of working with oai and anthropic models has been the silent failures and fallbacks when doing deep infra, cybersec, or RE.
It's not perfect, but worth trying out if this sounds familiar.
Thanks @edenchan for letting me know when this dropped
@ella_schlags Email their privacy team for a PIPEDA access request. It's annoying for them to do. You'll get a call from someone w authority, they will resolve in exchange for not having to continue with the request.
505 Followers 4K Followingж There’s no common name for your kind, because of how you remind with broken parts. ж 🏴☠️ ж {identify: null}ж mostly self taught ж 9r0d1c* ж PWSH
1K Followers 7K FollowingScala expert, AI for coding, data platforms, metadata management, process engineering, TDD. Creator @ScalaAlgorithms. Fan of digital and physical ergonomics.
0 Followers 239 FollowingLocal AI
My build: Gigabyte Radeon AI PRO R9700 32GB GPU, DEG2 eGPU, Thinkpad T14 with core ultra 7, Ubuntu 24.04, ROCm 7.12
Happy to answer any questions
217 Followers 432 Followingmaking models better one crud API at a time @cohere | previously a professional crudbob @amazon | cs + biochem @umich | views are my own
5K Followers 1K Followingbuilding box, the simplest, most affordable full VM sandboxes @asciidotdev w/ @luaroncrew
prev: HPC for black holes hunting @esa, gamedev, NN in rust, CAD AI
152 Followers 563 FollowingBuilding network-based computer use agents @ https://t.co/BuSn10E4gi. Systems engineer, database hugger, and a budding boulderer
5K Followers 883 FollowingFounder @tensorlake, building sandbox infra for agents. Past - Built Nomad @hashicorp, AI Infra @meta, Container Scheduler tech lead @linkedin and @netlfix