Skip to content

Web scrapingPlan

Collection that runs in real browsers.Thousands of them, behind one endpoint.

Cluster mode hands your code a single WebSocket endpoint and places launches across the workers behind it. The driver is the one you already use, over CDP. The browsers run on machines you provision, which is also the thing that decides what this costs to operate.

Get startedContact sales

Work like this usually starts and stays on a plan. Enterprise starts at $5,000 and is about the fleet, the seats and the people behind it rather than a larger profile count.

Fleet endpoint
cluster mode · one WebSocket endpoint
Drivers
Puppeteer · Playwright · Selenium
DNS
socks5h
Lifetime
one-time

01The work

The work

Three ways a run fails quietly

Loud failures are easy. These are the ones that report success and return nothing useful, which is the expensive kind.
  • Silent

    The response is fine and the content is not

    A page assembled client-side answers a plain HTTP client with a shell and no data. The run reports two hundred, the parser finds nothing, and the failure looks like a schema change rather than a missing engine.
  • The resolver

    The DNS query leaves from the wrong network

    A SOCKS5 proxy that resolves names locally sends the lookup from your resolver while the request arrives from the exit. Where the answer for a hostname is itself regional, the two disagree about where the client is.
  • Throughput

    Concurrency is a machine question, not a plan question

    How many browsers run at once follows the workers you provision. The scheduler places launches and retries one that fails on another worker; it does not conjure a machine, and no plan here sells you one.

02What carries it

What carries it

Four capabilities do the work here

Two of these exist on every platform in the category. The other two are why this page is not the same as everyone else's scraping page.
Your codePlaywright · Puppeteerone ws:// endpointworker-01browsers × nexit · residentialDNS resolves hereworker-02browsers × nexit · datacentreDNS resolves hereworker-03browsers × nexit · mobileDNS resolves hereconcurrency = workers × browserson machines you runsocks5h, so the name never resolves on your network
Fig. 01One endpoint, three workers, three exits. Concurrency follows the machines you run, and with socks5h the hostname is resolved at the exit rather than on your network.

Capabilities doing the work

Cluster mode
One endpoint in front of the whole fleetYour code holds one WebSocket connection. Launches are placed across registered workers and a failed launch is retried on another, so adding capacity is a machine you register rather than a client you rewrite.
SOCKS5H routingOnly here
The name resolves where the request exitsOne proxy per profile over HTTP, HTTPS, SOCKS5 or SOCKS5H. With socks5h the DNS query is resolved at the exit, which is what keeps hostname resolution and the request itself in the same place.
One-time profiles
A clean environment per run, when that is what you wantA one-time profile discards its storage when the session ends, so the next run starts from nothing. A persistent one keeps it, which is what collection behind a login needs. The lifetime is a property of the profile rather than a cleanup step in your harness.
Automation surface
The driver you have, plus a REST API and a typed SDKCDP for Puppeteer, Playwright and Selenium, a REST API for profile and session work, and a typed SDK over it. Deletion takes a whole set in one call; creation and duplication are one call each.

The whole fleet can run on infrastructure you own, which is the part of this that is not somebody else's per-gigabyte bill. Exit-derived timezone and locale are here too, and are table stakes in this category rather than a differentiator.

03Plan or enterprise

Plan or enterprise

Starts on a plan, becomes enterprise at fleet scale

The line is not request volume. It is how many machines you are operating and who is on call for them.
PlansTypical herePick one and pay by card
One machine, a few dozen browsersA worker on hardware you already have, driven from a script. This is the right start and frequently the right end.
EnterpriseFrom $5,000
Many machines, with a fleet manager in front of themCluster mode across hosts, validated configuration swaps and drain-then-roll updates belong to a deployment somebody operates, with seats for the people operating it.

This platform does not supply proxies. What bounds a run is the machines you provision and the exits you bring to them.

Plans and enterprise run the same platform. The difference is who operates the fleet and how many people need a seat on it.

Get startedPricing

04Where it stops

Where it stops

Two things this page will not claim

Both are shapes of the product as it is today rather than positions on a roadmap slide.
  • Not included

    Proxies are yours to bring

    Per-profile routing accepts HTTP, HTTPS, SOCKS5 and SOCKS5H endpoints that you supply. There is no bundled address pool, no residential inventory and no per-gigabyte resale.
  • One call each

    Bulk creation is a loop

    Deleting profiles takes a list of ids in one call. Creating and duplicating them do not, so provisioning ten thousand profiles is ten thousand calls against the create endpoint.

Point one driver at one endpoint

The first test worth running is a single worker, one socks5h exit and a page you already know is rendered client-side. If the driver connects and the DNS answer comes from the exit, everything after that is machines.

Endpoint
one WebSocket endpoint in front of every registered worker
DNS
resolved at the exit whenever the route is socks5h
Bounded by
the machines you provision and the exits you bring

The cluster mode documentation covers the endpoint, worker registration and retry behaviour in the detail this page deliberately does not.