Company Jobs API: every open role behind any domain

Paste company domains. Get every open job, in one schema, with salary when the company publishes it.

No proxies. No cookies. No login. The boards this actor reads are the companies' own applicant systems, not job aggregators - so a run answers in seconds and does not fall over when Indeed changes its markup for the third time this month.

What most job scrapers do, and why this is different

The usual offer	Here
Scrapes one site; LinkedIn or Indeed blocks it and the run returns nothing	Reads six ATS APIs directly - Greenhouse, Ashby, Lever, Workable, SmartRecruiters, Recruitee - and detects which one a company uses from its domain
Silent failure: no rows, no reason	Every company is reported: the ATS it uses, jobs on the board, jobs delivered, or the exact reason it could not be resolved (COMPANY_REPORT in run storage)
You pay per run, even when it returns an empty list	You pay per job delivered. A company with no open roles costs nothing
Descriptions, salary and remote flag arrive inconsistently	One flat schema for all six sources; salary parsed to min/max/currency/period
Needs residential proxies, billed to you	Zero proxy traffic - the sources are public JSON
Same list checked hourly = every board hit hourly, forever	Answers are cached in your own storage for 60 minutes (cacheMinutes): a repeat check inside that window is instant and sends no duplicate request

How it finds the board

A domain almost never matches the board slug, so detection tries, in order:

1. The careers page. example.com/careers and friends are read, and any ATS link on them is followed (boards.greenhouse.io/acme, jobs.ashbyhq.com/acme, and so on).
2. The domain itself as a slug against all six APIs.
3. If nothing answers, the company is marked NO_ATS_FOUND with a hint - pass the board slug (leverdemo) or the full careers URL instead of guessing. If the careers page names a provider this actor cannot read (personio today), the row says UNSUPPORTED_ATS and names it, so a found board is never reported as missing.

You can also pass a board slug or a careers URL directly; both are understood.

Input

{
  "companies": ["stripe.com", "figma.com", "notion.so", "canva.com"],
  "keywords": "engineer, data",
  "location": "remote",
  "onlyRemote": false,
  "includeDescription": false,
  "maxJobsPerCompany": 0,
  "maxCompanies": 50,
  "cacheMinutes": 60
}

• keywords filters on the job title, case-insensitive, comma separated. Filtered-out jobs are not charged.
• location is a substring match on the job location.
• includeDescription adds the full description as plain text; the price per job does not change.
• cacheMinutes reuses a result from your own key-value store within that window. 0 = always live. A company we could not resolve is cached for half the window, so a board that was briefly down recovers on the next run.

Output

One row per job:

{
  "company": "Stripe",
  "domain": "stripe.com",
  "ats": "greenhouse",
  "jobId": "4567890",
  "jobTitle": "Software Engineer, Payments",
  "department": "Engineering",
  "location": "London, UK",
  "remote": false,
  "employmentType": null,
  "salaryMin": null,
  "salaryMax": null,
  "salaryCurrency": null,
  "salaryPeriod": null,
  "postedAt": "2026-10-28T09:12:41-04:00",
  "url": "https://stripe.com/jobs/search?gh_jid=4567890",
  "status": "FOUND",
  "fromCache": false,
  "fetchedAt": "2026-11-04T15:30:00+00:00"
}

fromCache: true means the row was served from your cache - same data, no extra request to the board. The run summary also reports cacheHits and cacheMisses, so you can see exactly what was reused.

Who uses this

• Recruiters and sourcing teams - a target-company list becomes a live vacancy list.
• B2B sales - a company that is hiring three sales reps has budget; that is the strongest intent signal there is.
• Job boards and aggregators - fill categories from company boards instead of scraping aggregators.
• Market research - salary ranges and remote share by company, straight from the source.

Price

$0.0015 per delivered job ($1.50 per 1,000). A run that finds nothing costs nothing. Extraction of the description costs the same as the row itself.

The platform puts a default spending cap of about $4.63 on a run when the buyer does not set one, which is roughly 3,000 delivered jobs: a list of 50 companies on large boards reaches it. That cap is the buyer's setting, not ours - raise maxTotalChargeUsd in the run options when the whole list must be covered. When it is hit, the run stops and every company it never reached comes back as NOT_CHECKED_CAP with the reason.

Cost of a run

The actor only makes HTTP calls to public JSON endpoints: a 50-company run typically takes well under a minute on the smallest memory setting. With the cache on, a repeat check of an unchanged list makes almost no requests at all.

Notes

• Only boards the company published publicly are read; nothing is logged in to.
• Salary is returned when the ATS exposes it (Ashby and Lever usually do; Greenhouse usually does not).
• Workday, SuccessFactors and Taleo boards are not covered yet - tell us if your list needs them.
• maxJobsPerCompany is a per-company safety stop. When it keeps jobs back, the summary reports jobsOverMaxJobsPerCompany, so a capped board is never presented as a board that small.
• Two input rows can resolve to the same board (stripe.com and https://stripe.com/careers): the second answer is not thrown away. It goes through the same job-id gate, so a job already delivered is skipped and a job only the second answer carried is delivered and charged once. When it brings nothing new it comes back as DUPLICATE_BOARD with the name of the row that carries the jobs, and nothing is charged twice. Those extra jobs are counted as jobsFromDuplicateAnswers, because jobsMatched counts a board once.
• A company whose jobs the spending cap interrupted mid-way is marked partial: true in COMPANY_REPORT with the counts it did deliver - half a board is never reported as a whole one.
• maxCompanies defaults to 50: a longer list is cut at the input, and the companies beyond the limit come back as NOT_CHECKED_MAX_COMPANIES records with the reason (rows in the dataset when includeEmptyCompanies is true). companiesAsked counts distinct companies and companiesOverMaxCompanies counts what the cut took, so a 100-company list never reads as a 50-company answer. Case-insensitive repeats of one input (a.com, A.COM) collapse into one company and companiesDeduplicated says how many lines that was.
• If the run stops at your spending limit, the companies it never reached are still in COMPANY_REPORT with status: NOT_CHECKED_CAP and a reason, and the summary counts them as companiesNotChecked. A company that was not checked is never presented as a company with no open jobs.
• If a charge call fails twice in a row, the run stops instead of handing out the rest of the list for free: the summary says billingActive: false with stopReason: billing-error, and the companies it never reached come back as NOT_CHECKED_BILLING. The row that was already delivered stays delivered and stays readable, but it is not paid for: rows handed out while billing was failing are free to the buyer.
• If the platform reports that this run is not pay-per-event (for example the price is not set up yet), nothing is charged at all: the rows are still delivered, and the summary says billingActive: false with stopReason: billing-not-active. A buyer is never told he paid for rows that came out free.

Cache and billing

A row served from the cache is charged like any other row: you pay for delivered jobs, not for network calls. What the cache buys you is time - a repeated company resolves in seconds instead of probing six ATS APIs again.