Skip to content

Menu

Budget residential network requests

Budget residential HTTP and browser fetches with current credit rates, a tested offline calculator and explicit page-size and routing scenarios.

By Codex. Checked 2026-11-15.

A network allowance of 12 credits can cover 240 standard residential HTTP fetches at the current rate. It covers only 60 residential browser fetches. Before starting a job, estimate how many pages will need each route and check how much the result changes if more pages need rendering.

The example below budgets 100 successful paid fetches. It needs Node 22 and four downloadable files. The calculator runs offline using a saved pricing response, so you can test the budget before making an authenticated request.

Read the prices in the right units

Scraipe’s public pricing response labels its network prices as millicredits. Divide by 1,000 to express them in credits. The snapshot checked for this guide gives the following rates. Current pricing response

Successful residential fetchPrice in the responseCredits per fetch
HTTP, up to 262,144 wire bytes500.05
HTTP, above 262,144 and up to 1,048,576 wire bytes1500.15
Browser, within its 1,048,576-byte budget2000.20

The API refuses an HTTP page beyond its wire limit with too_heavy. Direct, cached and conditional responses have zero network price under this contract. The pricing response states that failed requests are uncharged. That does not make a blocked page a successful result you can include in your dataset. Track completion alongside spend. Network options and limits

For extraction, meta.credits.compile is a separate charge component. Add learning, repair and fallback costs when preparing the total job allowance. A previously learned extractor can still need a paid fetch on its next visit. Estimate a complete extraction workload

Check the route before budgeting it

Use egress: "direct" when the job should use Scraipe’s servers only. Use auto when the routing policy may use residential access, or residential to start on that network. Crawl defaults to direct access. Other operations can have different defaults, so include the intended policy in your saved request.

An API key with the read scope can request a quote. Replace this illustrative host with a permitted page and include the fields you intend to extract.

bash

curl --fail-with-body -sS \
  https://scraipe-fc66a552921e.herokuapp.com/v1/quote \
  -H "Authorization: Bearer ${SCRAIPE_API_KEY}" \
  -H 'Content-Type: application/json' \
  --data '{"url":"https://catalog.example/item/42","fields":["title"],"egress":"auto"}'

A quote consults known routing information without fetching the page. Check requiresPurchase, any refused reason, and the estimated compile and egress components. The quote’s max can include learning as well as network use. Keep that combined figure out of a network-only unit price. A quote also leaves the actual result, page size and service availability to be established by a real permitted request. Quote contract

We reproduced five quote cases through the local HTTP route harness. A standard residential case quoted 0.05 network credits, a heavy case quoted 0.15, and a forced residential request quoted 0.05 even for a host known to work directly. The unpaid test account returned requiresPurchase: true. Choosing direct access quoted zero network credits. These observations used simulated accounts, hosts and transport.

Run the network allowance calculator

Download the calculator, sample workload, saved pricing response and tests into one directory. Run the documented sample with the included snapshot.

bash

node --test budget.test.mjs
node budget.mjs plan.json pricing.json

The sample’s 100 successful paid fetches comprise 60 standard HTTP pages, 20 heavy HTTP pages and 20 browser pages. Their costs are 3, 3 and 4 credits respectively, totalling 10 credits. The 12-credit network allowance leaves 2 credits of headroom.

The allBrowser scenario prices the same 100 fetches at 0.20 each. It totals 20 credits, leaving a shortfall of 8 credits. The output also shows affordableBrowserFetches: 60. Those are scenarios for the successful paid fetches you entered. Additional visits or newly discovered pages add work to the budget.

For a smaller change, move 20 fetches from httpStandard to httpHeavy in the sample. The mix then costs exactly 12 credits. This shows how a change in response size alone can consume the headroom, even when page count and rendering remain stable.

The calculator exits with status 0 when it produces a report, including a report with withinAllowance: false. It exits with status 2 for invalid inputs. Removing the browser count fails validation. The script requires an explicit zero when you expect no browser fetches.

Calibrate the counts with a pilot

For your own job, group URLs by host, page type, country and extraction requirements. Run a small representative pilot once the account and service can use the required route. Record the actual meta.egress.mode, meta.egress.priceClass, meta.credits.egress, success or failure, and request ID. Use those observations to revise the mix. Include a fresh successful fetch again when the job revisits a URL. Count a stored idempotent replay once with its original operation.

Download current prices into a separate file before planning another run.

bash

curl --fail-with-body -sS \
  https://scraipe-fc66a552921e.herokuapp.com/v1/pricing \
  -o pricing-current.json
node budget.mjs plan.json pricing-current.json

Preserve the snapshot alongside your assumptions. If it is incomplete, has changed units or introduces different browser prices by size, the calculator asks you to revise the model. An estimate from a small pilot can miss a layout or size class. Include a browser-heavy scenario when the mix is uncertain, and revisit it after the first batch. Read actual request charges

Turn the allowance into a run decision

The sample can proceed under its ten-credit mix with two credits available for variation. If the job needs room for 100 browser fetches, allocate 20 network credits or reduce that scenario’s page count to 60 within the existing allowance. Investigate incomplete results before treating low spend as a saving.

The calculator produces a planning report. To stop work based on spend, use your scheduler to check recorded charges before dispatching the next bounded batch, accounting for requests already in flight. The v1 contract lists maxCredits, maxJobCredits and egress.maxCredits among fields it does not read. Sending one of those fields will not enforce this allowance. Set explicit network policy and finite job scope, then reconcile actual spend after each batch. API contract

Sources and verification

Try the next step

Try the playground

Continue building