Stan Sadokov | NodeMaven

Building @NodeMaven, proxy infrastructure for web/data engineers and AI agents. Writing about scraping systems, IP trust and bootstrapping.

[Stan Sadokov | NodeMaven @StanSadokov]

I’m Stan. I build @nodemaven , proxy infrastructure for people running scraping, browser automation and AI agents in prod.

I write about what breaks between the demo and the ops: IP quality, sessions, browser identity, reliability and cost.

[Stan Sadokov | NodeMaven @StanSadokov]

A whole market sells one promise, you never touch a scraper
one vendor's homepage shows 10 billion records delivered, another counts 15,178 customers

here is the map for august 2026
who you call when you want the data and not the scraper

managed extraction, you send a spec and datasets come back
@scrapehero, @grepsr, @promptcloud, @DatamamScraping, ficstar, xbyte enterprise crawling, @actowizsolution, @crawl_now, autoscraping , @Prowebscraper

enterprise data on demand
@zytedata, BrightData, @Oxylabs_io, @nimble_search

ai-native extraction, the model figures out the page instead of a human writing selectors
@diffbot, @riveterhq, kadoa

the marketplace layer
@DataradeHQ, aws data exchange, @Snowflake marketplace

the product in every provider row is the same
somebody else runs the scraper, watches the quality and eats the anti-bot problem

even if you never touch the scraper, somebody still buys the addresses it runs through
that layer is what we provide at @nodemaven

for those of you running scrapers in production
what would you never outsource, and what are you happy to just buy?

[Stan Sadokov | NodeMaven @StanSadokov]

repos
scrapling: https://t.co/ctVU6mk48T
patchright: https://t.co/HrkapxeUTs
camoufox: https://t.co/0NaA82jeOd
pydoll: https://t.co/W8WnZSd6lC
google-maps-scraper: https://t.co/ihrQUbIIM4

[Stan Sadokov | NodeMaven @StanSadokov]

Scrapling has 75,6k github stars
its stealth mode runs on patchright, which has 4k

5 open source scraping tools, tested and validated by our team

@Scrapling_dev, by @D4Vinci1
the adaptive parser re-finds elements after a site redesign, so markup changes do not always mean rewriting selectors
catch: the stealth engine switched from camoufox to patchright at 0.3.13, older write-ups describe a different browser

patchright
change the import and the rest of your playwright code stays as it is, with the automation tells patched out
catch: chromium only, and it disables the console api. anything you built on console output stops working

camoufox
fingerprints spoofed inside the firefox c++ layer, where javascript cannot reach them
catch: it is a patched firefox fork. firefox is less commonly used by regular web users, so anti-bots always score it with less trust

pydoll
chrome over cdp with no webdriver flag anywhere
catch: the captcha feature is not a solver. the docs say so plainly, it automates the same click a person would make

google-maps-scraper
self-hosted maps extraction in go with a rest api
catch: fast mode is still beta and returns up to 21 results per query. wider coverage means grid mode across a bounding box

all of that is fixable in code. a bad ip is not, which is why you still need decent residential proxies
we run those at @nodemaven

which of these are you running, and what broke first?

[Stan Sadokov | NodeMaven @StanSadokov]

i spent 10 years on anti-detection
first @multilogin_en on the browser side, then @nodemaven on the ip side

we finally shipped the two as one product
scraping browser, a cloud browser you talk to over cdp

you generate a cdp url and point playwright or puppeteer at it
the scraping logic stays yours, the environment it runs in becomes our job

every session starts with a managed fingerprint, automatic captcha solving and a clean residential or mobile ip

in the dashboard
templates for e-commerce, b2b enrichment, real estate and price monitoring
ai prompts that return structured json
live browser and session recordings to debug runs
profiles that keep cookies and local storage
up to 50 parallel sessions

live on every rotating residential plan, start with a template

[Stan Sadokov | NodeMaven @StanSadokov]

https://t.co/FkKwhx4lPz
agpl-3.0, macos and windows

[Stan Sadokov | NodeMaven @StanSadokov]

nextbrowser is alive
three months ago i wrote that it was dead

the cloud version still is
we closed it for very boring reasons
sessions ate 4 to 8 gigabytes of ram each, the codebase needed a full rewrite, and i did not see a market big enough to justify both

so we stopped
i think that was the right decision

the team did not stop
they rewrote it from scratch, 47 releases since the beginning of july, and they were right

so we opened it
agpl, every line public, free

now it is a desktop app that runs local ai agents in managed browser sessions
profiles, proxies, fingerprint rotation, schedules, live view in one place

in may i wrote that every closed product is a tuition payment
this is what the tuition bought

repo in the first reply

[Stan Sadokov | NodeMaven @StanSadokov]

repos
curl_cffi: https://t.co/75ChdzsvC9
fingerprint-suite: https://t.co/1YB2Yi3FAn
obscura: https://t.co/9RZeBhyj02
trawl: https://t.co/Zm2nl2rbaT
stealth-browser-mcp: https://t.co/rA5vs8tKBn

[Stan Sadokov | NodeMaven @StanSadokov]

Five open source anti-bot repos, 32k github stars between them
tested by our team, here is what each one covers

a public page sees your ip and asn, then tls, then http/2. canvas and webgl matter only once a script runs

you install a second stealth library and nothing changes
both cover the same layer, and the block is somewhere else

curl_cffi
impersonates browser tls and http/2 fingerprints from python
runs no javascript, so js challenges are where it stops

fingerprint-suite
@apify toolkit for node
one call gives a playwright context where headers and js apis come from one generated fingerprint

obscura
independent rust and v8 engine without chromium, 21k stars at v0.2.0
rendering and web api behavior can differ from chromium

trawl
escalates from plain http to a cached session, then a fresh browser solve, then a residential proxy
if the early steps work it never reaches the proxy

stealth-browser-mcp by @ImVibhek
a stealth browser over mcp, pinned to nodriver 0.47.0
you get that driver's fixes only when someone bumps the pin

start with curl_cffi when you do not need page javascript
it is the cheapest thing to try

check the recent commits before you install any of them 👇

[Stan Sadokov | NodeMaven @StanSadokov]

@OpenAI turned off Atlas on August 9, 9 months after launch, and keeps the agent side alive inside ChatGPT, including a cloud browser on its own servers.

Here is the map for August 2026: who runs the browser when your agent has to act on a page instead of reading it.

The Atlas category: @comet by Perplexity, @FellouAI, @strawberry, @genspark_ai

Cloud browser runtimes: @browserbase, @AnchorBrowser, @steeldotdev, @usekernel, @hyperbrowser, @browserless, @lightpanda_io

Open source agent frameworks: @browser_use, @Stagehanddev, @skyvernai, @nottecore

No-code task agents: @airtop, @browseract, @BrowseAI, @browserbots

The primitives all of them sit on: @playwrightweb, Puppeteer, CDP

On September 15 Cloudflare changes the defaults for new domains: the Agent category gets blocked on pages that display ads.

@Cloudflare defines that category as automated activity acting in real time on a person's behalf. That definition covers every tool above.

Each of those sessions ends at the same gate: an IP and a fingerprint the site judges as one unit, so one burned address kills the run no matter which of them started it.

Those of you that run agents or scraping in production: which of these is in your stack?

[Stan Sadokov | NodeMaven @StanSadokov]

@skyzer4ever @Unigox_global sic story! some day I wanna grow to become an enterprenuer like you boss

[Stan Sadokov | NodeMaven @StanSadokov]

gpt-5.6 just went public 🔥

the loudest number is on the pricing page
the intelligence you paid full price for in june now costs half

three models at once

sol for hard agentic work
terra for gpt-5.5 level output at half the price
luna for volume

what builders will feel first

ultra mode - sol spawns subagents and runs parts in parallel
speed - sol on cerebras at up to 750 tokens per second
gpt-live - full duplex voice that listens while it speaks and jumps in on its own

voice stops being a walkie-talkie
it becomes a real conversation

cheaper faster and now it talks
the web this lands on did not change
and that gap is where our whole industry lives

what are you pointing sol at first?
@ChatGPTapp

[Stan Sadokov | NodeMaven @StanSadokov]

@rbatista19 @Cloudflare @zytedata @apify @BrightData @Oxylabs_io we indeed are! version 2 coming up soon :)

[Stan Sadokov | NodeMaven @StanSadokov]

@Cloudflare blocked over 416 billion ai bot requests
and the industry that still gets the data out just produced a 275 million dollar exit

here is the map for july 2026

full stack platforms
@zytedata , @apify , @BrightData , @Oxylabs_io

scraping api specialists
@ScraperAPI , @ScrapingBee , scrapfly, https://t.co/ziMAIaSgAa , @ZenRowsHQ , @scrapingdog , crawlbase

ai native extraction
@firecrawl , @scrapegraphai , @JinaAI_

search apis for agents
@tavilyai , exa , @p0 , @serp_api , serper , @dataforseo

open source
scrapy , @crawl4ai , @crawleecloud , scrapling

the money went to the search layer
@tavilyai sold for 275m to @nebiusai
parallel raised at around 2b valuation

every request still exits through the same gate
an ip and a browser fingerprint the site judges as one unit

one burned ip kills the run no matter which tool sent it

that gate is where we work
@nodemaven filters the addresses before they reach your scraper

those of you running scraping in production
which of these tools is in your stack

[Stan Sadokov | NodeMaven @StanSadokov]

@rbatista19 great article! seems this Verify wall is starting to massively affect scrapers. did you manage to get the JS payload? our team is looking into it too. if you have the payload - we can help you analyze it and share results what happens under the hood.

[Stan Sadokov | NodeMaven @StanSadokov]

the ip was in berlin
the browser timezone said chicago
the run died on the first request
and of course the proxy got the blame

i see this every week

in most cases the ip was fine
the person just ran a mismatched browser profile

sites read everything at once

ip geolocation
accept-language
timezone from javascript
webrtc

together they describe a person that does not exist
anti-bot systems catch exactly that

years looking at fingerprints at @multilogin_en and ips at @nodemaven taught me one thing

browser environment
infrastructure
and ip
all have to agree

an expensive proxy will not fix a sloppy profile
a perfect profile will not fix burned ips

at @nodemaven we remove that last variable
every ip passes our quality filter before it reaches you

before you launch a profile check at least these

timezone vs exit ip country
locale vs exit ip country
webrtc for leaks
geolocation api for the same mismatch

takes about two minutes
if anything disagrees with the ip fix that first

scrapers what got you blocked more often
bad ips or bad profiles

detection builders which mismatch is the loudest tell right now

[Stan Sadokov | NodeMaven @StanSadokov]

One of the best open source repos for web scraping of all times

[Massi — oss/acc @0xMassi]

Four months ago I open-sourced webclaw. Today it just passed 2k GitHub stars

One founder. 57 releases. No funding, no ads. Developers spread it.

Turns any URL into clean Markdown or JSON for AI agents.

Rust, AGPL, self-hostable.

[Massi — oss/acc @0xMassi]

Four months ago I open-sourced webclaw. Today it just passed 2k GitHub stars

One founder. 57 releases. No funding, no ads. Developers spread it.

Turns any URL into clean Markdown or JSON for AI agents.

Rust, AGPL, self-hostable.

[Stan Sadokov | NodeMaven @StanSadokov]

@0xMassi I have been following your journey for some time man, and its nothing short of amazing! To 10k stars next!

[Stan Sadokov | NodeMaven @StanSadokov]

two days of q3 planning at nodemaven

i spoke maybe 17% of the time

that’s what i’m proudest of

as ceo my goal is simple

speak as little as possible

negotiate achievable objectives with the team

leave the room only with shared understanding

every idea gets roasted including mine

product challenges marketing. marketing storms product. sales pushes back

in hybrid meetings the key rule

remote people must be able to interrupt

if the loudest voice in the room sets the pace you lose half the brains

we spent most time on the bigger picture

where the internet goes in the age of ai

clean verified data is becoming the rarest thing

nodemaven wants to be the layer that provides it

big things are coming

founders what’s your one rule for great hybrid meetings