multi-account-isolation

Multi-Account Isolation

Keeping several of your own accounts from being tied together by a platform is not one problem, it is a stack of them. A perfect browser fingerprint fixes exactly one layer. Accounts far more often get linked by an IP, a shared cookie, a mismatched clock, or the same recovery phone number than by a canvas hash.

This skill is the checklist. For the code that implements the browser layer, see the anti-detect-browser skill.

Scope. This is about accounts you own, or accounts you operate with the holder's authorization - an agency running client profiles, a seller with several storefronts, a team whose ad accounts must not be cross-attributed. It is not about getting into accounts that are not yours, manufacturing fake accounts or engagement, or coming back from a ban you earned; see Acceptable use. Proxy URLs and API keys in the examples always come from the environment.

The model: platforms link on many layers, and only need one

Linkage systems build a graph. Every account is a node; every shared signal is an edge. One strong edge (same payment card) or several weak ones (same /24 subnet + same screen resolution + same active hours) is enough to merge two nodes into "one operator".

So the goal is not "beat the fingerprint test". It is leave no edge. Getting nine layers right and the tenth wrong still merges the nodes.

Linkage surface

The bottom four rows are not solved by an anti-detect browser. If accounts keep getting linked with the browser layer correct, look there first.

The rule: one account, one of everything

account  →  profile  →  fingerprint  →  proxy  →  timezone  →  identity data
   1     :     1      :       1        :    1    :     1      :        1

Any shared cell in that row is a potential edge. Profiles are unlimited and free on every antibrow plan, so there is never a reason to reuse one.

Setup

Pair each account with its own profile and its own sticky proxy. The persona is drawn once and frozen, so the account sees the same device on every later launch - which is what a real user looks like.

import { AntiDetectBrowser } from 'anti-detect-browser'

const ab = new AntiDetectBrowser({ key: process.env.ANTI_DETECT_BROWSER_KEY })

const accounts = [
  { profile: 'shop-us-01', proxy: process.env.PROXY_US_1, tags: ['Windows 10', 'Chrome'] },
  { profile: 'shop-us-02', proxy: process.env.PROXY_US_2, tags: ['Apple Mac', 'Safari'] },
  { profile: 'shop-de-01', proxy: process.env.PROXY_DE_1, tags: ['Windows 10', 'Edge'] },
]

for (const a of accounts) {
  const { browser, page } = await ab.launch({
    profile: a.profile,               // isolated cookies, storage, login state
    proxy: a.proxy,                   // sticky, one per account
    fingerprint: { tags: a.tags },    // frozen at first launch, replayed after
    label: a.profile,                 // floating label so windows are tellable apart
  })
  // ... work this account, then close it ...
  await browser.close()
}

Python, same profile format and same on-disk identity:

import os
from antibrow import launch

with launch(
    profile="shop-us-01",
    proxy=os.environ["PROXY_US_1"],   # sticky proxy URL from the environment, never a literal
    geoip=True,            # timezone + WebRTC follow the proxy exit
    label="shop-us-01",
) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    print(browser.timezone, browser.public_ip)

Proxy selection

• Sticky, not rotating. A logged-in session whose IP changes mid-flight looks like a hijacked session. Rotation belongs in scraping, not in account operation.
• Residential or mobile for consumer platforms. Datacenter ranges are widely tagged.
• Match the account's claimed location to the proxy, and keep it stable. An account that has always been in Ohio and suddenly appears in Vietnam is a stronger signal than any fingerprint.
• Never share one exit across accounts, and do not reuse the exit of an account that was already banned.

Fingerprint variety

Do not give every account ['Windows 10', 'Chrome']. Real populations are mixed. Vary the tags across the fleet - Apple Mac, Android, Edge, Mobile - and let each account keep its own draw forever.

Warm-up

New accounts that immediately do the thing you made them for are the easiest to catch. Age each account with ordinary behaviour before it does anything of value: browse, read, follow a few unrelated things, come back the next day. Days, not minutes. Warm-up is a behaviour problem, so no browser feature substitutes for it.

Verify before you trust the setup

Run a fresh profile against a detection suite through its own proxy and confirm the layers agree:

• CreepJS - cross-layer consistency, worker vs main thread
• whoer.net - IP, timezone and locale agreement at a glance
• browserleaks.com/webrtc - real IP leaking past the proxy
• pixelscan.net - IP-to-fingerprint coherence
• npx liarjs (liarjs.dev) - ~40 open-source consistency rules, runnable in CI

Check specifically: does the reported timezone match the exit IP's country, does the WebRTC candidate show only the proxy, and does the canvas hash stay the same across two launches of the same profile (a changing hash is itself a flag).

Troubleshooting: accounts still getting linked

Work down in this order - the cheap layers first, because they are the ones usually at fault:

1. Two accounts in one profile? Check that no profile name was reused. list_profiles or ~/.anti-detect-browser/.
2. Shared or recycled IP? Confirm each account's public_ip is distinct and that none belonged to a banned account.
3. Timezone mismatch? Print browser.timezone and browser.public_ip and confirm they agree.
4. Shared account metadata? Recovery email, phone, payout account, billing address - the most common real cause.
5. Behavioural overlap? Same active hours, same content, accounts interacting with each other.
6. Only then suspect the fingerprint - and verify it with the tools above rather than assuming.

What this cannot do

• It does not make a banned account come back.
• It does not hide a shared payment instrument or a shared payout account from a platform that checks them.
• It does not fix content or behaviour that a platform would penalise anyway.
• It does not defeat identity verification - a document check is not a fingerprint problem.

What the runtime touches, and how to check it

Running accounts through any tool means handing that tool session cookies and proxy credentials, so it is fair to ask what the runtime does with them. For antibrow:

The kernel is a closed-source Chromium build - that is the tradeoff for spoofing living in C++ rather than in an injectable script - so verify behaviour rather than take it on faith:

python -m antibrow info          # kernels, profiles, license state, cache dir

browser.plan.redacted_args()     # exact kernel command line, secrets masked - safe to paste in a bug report

Point the whole thing at a proxy you can read logs on, or at a local MITM proxy, and watch what leaves the machine during a launch and a browsing session. Pin the SDK version and check the published hash (npm view anti-detect-browser@2.2.0 dist.integrity) so the code being audited is the code that runs. If a deployment must not phone home at all, this tool is the wrong choice: license verification is compiled into the kernel and there is no offline mode.

Acceptable use

Managing your own accounts across platforms, running client accounts with authorization, testing your own anti-fraud stack, and scraping public data are the intended uses. Bulk account-creation abuse, credential stuffing, evading a ban you were given for a policy violation, and automating systems you have no authorization to access are out of scope. Complying with the terms of the platforms being used is the operator's responsibility.

Related Skills

• anti-detect-browser - the SDK, profiles, fingerprints, proxies, and REST API that implement the browser layer
• browser-mcp-agent - MCP server mode, for letting an AI agent drive an isolated profile itself