AI browser agent
Residential proxies in browser-use
browser-use drives a real Chromium for an LLM agent. Sites see the machine it runs on, usually a cloud IP that gets challenged. Pass a residential proxy in ProxySettings and the agent browses from a home connection in the country you pick.
- Host
- via.proxies.vision
- Ports
- 9000 rotating · 10000 sticky
- Same IP for
- up to 24 hours
- Price
- $1/GB, never expire
The code
browser-use takes the proxy as ProxySettings on the Browser. Replace USER and PASS with your login from the dashboard.
from browser_use import Agent, Browser, ChatBrowserUse
from browser_use.browser import ProxySettings
browser = Browser(
proxy=ProxySettings(
server="http://via.proxies.vision:10000",
username="USER-country-us-session-agent1-lifetime-30",
password="PASS",
),
)
agent = Agent(
task="Open example.com and read the headline",
llm=ChatBrowserUse(),
browser=browser,
)Set it up in browser-use
As in browser-use's documentation today.
01
Get your login
Sign up and copy your proxy username and password from the dashboard.
02
Pass ProxySettings
Give the Browser a ProxySettings with our server, your username and your password, as in the example.
03
Keep one IP per task
A session in the username keeps the agent on one home IP while it logs in and clicks through; use a new session name per task.
04
Check the exit
Let the first task open an IP checker. It should show the country from your username.
Questions
- Rotating or sticky?
- Port 9000 gives every new connection a new home IP, good for crawling many pages. For a logged-in flow that has to keep one IP, put a session in the username (USER-session-job1-lifetime-30) and use port 10000; the same session name keeps the same exit for its lifetime.
- Does it work with a remote browser (cdp_url)?
- Yes. browser-use documents ProxySettings next to cdp_url, so the remote Chrome connects through the proxy.
- What does it cost?
- $1 per GB of traffic, whatever the country, city or network. No subscription, and unused bandwidth never expires. An agent that loads full pages uses more than one fetching HTML, so block images and media where the task allows it.
