ATS Jobs Feed – Greenhouse, Lever, Ashby & Workable Job Listings

Open jobs for a list of companies, read straight from each company's public applicant-tracking-system job board. One dataset row per job with title, company, location, remote flag, salary when the employer publishes it, department, posted date, and the job URL. Turn on newSinceLastRun and a scheduled task becomes a feed of only the jobs it has not shown you before.

The Actor calls the official public job-board APIs that these ATS vendors publish for career pages. It does not scrape HTML, launch a browser, use a proxy, or solve CAPTCHAs. Requests run one at a time with a jittered 1–2 second pause.

What it does

Every endpoint below was checked with a live request on 2026-11-02. An unknown slug answered HTTP 404 on all four.

ATS	Endpoint called	Docs
Greenhouse	https://boards-api.greenhouse.io/v1/boards/{slug}/jobs?content=true&pay_transparency=true	Job Board API
Lever	https://api.lever.co/v0/postings/{slug}?mode=json	lever/postings-api
Lever (EU)	https://api.eu.lever.co/v0/postings/{slug}?mode=json	same README, EU instance links
Ashby	https://api.ashbyhq.com/posting-api/job-board/{slug}?includeCompensation=true	Public Job Posting API
Workable	https://apply.workable.com/api/v1/widget/accounts/{slug}	Not documented by Workable (see Limitations)

Notes from that check:

• Greenhouse documents pay_transparency=true on the single-job endpoint only. On 2026-11-02 the list endpoint accepted it too and returned pay_input_ranges on all 213 Airbnb jobs, so one request per board still gets structured pay. When the ranges are empty, the Actor looks for a "$120,000 – $150,000" pattern in the job content.
• Lever returns the whole board in one response unless limit is passed. The Actor does not page. createdAt is epoch milliseconds and is converted to ISO UTC.
• Ashby's documented job object has no id, but the live response includes one. The Actor uses id and falls back to jobUrl. Jobs with isListed: false are skipped because Ashby says they are direct-link only.
• Workable's widget feed is the JSON its embeddable careers widget calls. It answered 200 for huggingface and 404 with plain text for an unknown slug. It has no salary field.

Rows are filtered by location, remote, and keyword, then deduped on (ats, company slug, job id). The same job id at two companies stays two rows.

Input

fixtures/INPUT.json:

{
  "greenhouse": ["airbnb"],
  "lever": [],
  "leverEu": [],
  "ashby": [],
  "workable": [],
  "location": "",
  "remote": "any",
  "keywords": [],
  "newSinceLastRun": false,
  "stateStoreName": "ats-jobs-state",
  "maxItems": 3,
  "delaySeconds": 1,
  "maxSpendUsd": 1
}

This file is the CI fixture: one Greenhouse board and three jobs. Add Lever, Ashby, or Workable slugs for a wider run. keywords of engineer and engineering keep both word forms.

• Slugs are the last part of the public board URL: boards.greenhouse.io/airbnb, jobs.lever.co/leverdemo, jobs.ashbyhq.com/Ashby, apply.workable.com/huggingface. A full board URL is accepted and reduced to its slug. A jobs.eu.lever.co URL, or a name under leverEu, goes to Lever's EU API.
• location keeps a job when any of its locations contains the text (case-insensitive). Commas separate alternatives. The alternative Remote also matches any job whose ATS flags it as remote (Ashby isRemote, Workable telecommuting, a remote workplace type), even when the location text does not say "Remote".
• remote: any, remote (fully remote only), or exclude. Hybrid counts as not remote.
• keywords keeps a job whose title or department contains any of them as a whole word (case-insensitive). ai does not match "Maintainer" and engineer does not match "Engineering", so list both forms when you want both. A single comma-separated keyword string is accepted too.
• newSinceLastRun skips jobs pushed by an earlier run. Seen keys and their firstSeen time live in the named key-value store stateStoreName (default ats-jobs-state), which persists between runs. Use a different store name per saved task to keep feeds apart.
• maxItems is an exact cap on job rows pushed (default 100, maximum 10,000). Status rows do not count. When the cap is reached, later boards are not requested.
• maxSpendUsd defaults to 5 (minimum 0.01, maximum 100). A job row is pushed only when its charge fits under this cap. Once it does not, the job feed stops and one free spend_capped row says how many matching jobs were held back.

Output

One dataset item per job, plus a status row for each board that was unknown, disallowed, or failed, one blocked row per blocked host, and one spend_capped row if the spend cap ended the job feed. Key-value SUMMARY records counts (found, newJobs, repeatJobs, billedJobs, heldBackJobs, notFound, blocked, errors, mergedDuplicates, filteredOut, skippedSeen, skippedBlockedHost, truncated, chargeCapped), spend, and one report per board (ats, slug, status, fetched, kept).

{
  "ats": "greenhouse",
  "company": "Airbnb",
  "companySlug": "airbnb",
  "jobId": "1234567",
  "title": "AirCover Specialist, Claims Experience",
  "location": "Canada",
  "locations": ["Canada"],
  "remote": true,
  "workplaceType": "remote",
  "salary": {
    "min": 58000,
    "max": 68000,
    "currency": "CAD",
    "interval": "year",
    "raw": "Canada Annual Pay Range: CAD 58,000 - CAD 68,000"
  },
  "department": "Community Support",
  "postedDate": "2026-10-28T14:52:05+00:00",
  "url": "https://careers.airbnb.com/positions/1234567?gh_jid=1234567",
  "firstSeen": "2026-11-02T12:00:00+00:00",
  "sourceStatus": "found",
  "decision": "new",
  "ok": true,
  "error": "",
  "scrapedAt": "2026-11-02T12:00:00+00:00"
}

salary is null when the job lists nothing. min, max, currency, and interval are filled when the ATS publishes structured pay (Greenhouse pay ranges, Lever salaryRange, Ashby salary components) and are parsed from text otherwise. interval is one of year, month, week, day, hour, or null when the source does not say. postedDate is ISO UTC; Workable publishes a date only (YYYY-MM-DD). decision is new the first time a job is pushed and repeat after that (only when newSinceLastRun is off).

sourceStatus	Meaning	job event
found	A job was returned. Pushed only when maxSpendUsd covers its charge	yes
not_found	The ATS has no board with that slug (HTTP 404), or nothing matched the filters	no
blocked	HTTP 401, 403, or 429 after retries, or a challenge page instead of JSON. Later boards on that host are skipped; other hosts still run	no
spend_capped	maxSpendUsd or the platform charge limit ended the job feed; error says how many jobs were held back	no
robots_disallowed	robots.txt disallows the board URL	no
error	Bad slug, invalid JSON, a 5xx after retries, or robots.txt could not be read	no

Limitations

• Coverage is the four ATSs above. Workday, iCIMS, SmartRecruiters, and others are not read.
• A board is read in full in one request. Lever and Workable boards are not paged, and a very large board is capped at 32 MB.
• Workable does not document the widget endpoint. Its documented API needs an employer token. The widget feed can change without notice, and Workable sits behind Cloudflare, which may answer 429 from cloud IPs. That skips the remaining Workable boards with one blocked row.
• Remote is the ATS's own workplace field when it has one (Greenhouse "Workplace Type" metadata, Lever and Ashby workplaceType, Workable telecommuting), else the word "remote" in a location.
• Text salary parsing reads the first currency range it finds. A $ amount is reported as USD.
• newSinceLastRun state keeps up to 200,000 keys per store and drops the oldest first. Only jobs that were pushed and charged are recorded, so a job held back by maxItems or maxSpendUsd is still new next run.
• On a block the Actor pushes one blocked row for that host and skips its remaining boards. Boards on other ATS hosts still run, and the run exits 0. It does not retry through a proxy.

Robots and terms

robots.txt is fetched once per host and evaluated with scripts/lead_automation/crm_enrich_robots.py (longest match, User-agent: *). A 4xx robots answer means no rules (RFC 9309). A 5xx or network failure skips that host for the run. On 2026-11-02, boards-api.greenhouse.io disallowed only /embed/, api.lever.co and api.eu.lever.co allowed / with Crawl-delay 1, apply.workable.com allowed everything, and api.ashbyhq.com/robots.txt answered 401.

Greenhouse says job board data is publicly available with no auth on GET. Lever and Ashby publish these endpoints for customers' own career sites. Job content belongs to the employer. Use it to link back to the original posting.

Pricing

Pay per event from .actor/pay_per_event.json.

Event	Price	When it is charged
job (primary)	$0.002 (placeholder)	A job row is pushed

The $0.002 price is a placeholder pending a pricing decision. Apify Store, read on 2026-11-02:

• fantastic-jobs/career-site-job-listing-api: "from $4.00 / 1,000 jobs", pay per event, 1,649 monthly users, 4.11 stars.
• fantastic-jobs/greenhouse-jobs-api: "from $1.20 / 1,000 jobs".
• fantastic-jobs/lever-co-jobs-api: "from $0.80 / 1,000 jobs".

"From" is the lowest Store-discount tier. The undiscounted per-event price was not visible on those pages, so it is unverified. A third-party tracker (apifystats.com) lists the Career Site API at $12 / 1,000. That figure is not confirmed.

There are no proxy or captcha events because this Actor never uses either. not_found, blocked, robots_disallowed, spend_capped, and error rows are saved and not billed. Once maxSpendUsd cannot cover the next job, no further job rows are pushed. Dataset items are not a charge event. The platform still applies its start event. Memory default is 512 MB and the default timeout is 1,800 seconds.

Deviation from the template contract

The template keeps headless, maxProxyCostUsd, allowCaptcha, proxyConfiguration, and proxyGroups in every input schema. This Actor drops them. It reads public JSON APIs with plain HTTP, never launches a browser, and stops on a block instead of spending to get past it. That is the same stop-on-challenge path contractor-license takes. Only maxSpendUsd is kept.

Two further differences:

• The shared SpendLedger counts whole cents, so $0.002 would round to 0 and the cap would never stop charging. src/spend.py subclasses it as JobLedger in micro-dollars with the same can() / commit() contract, and push_charged_rows uses it.
• The Dockerfile uses apify/actor-python:3.12 instead of the shared Playwright image. It still leaves WORKDIR at /usr/src/app, copies only files inside this folder, and sets CJM_REPO_ROOT=/usr/src/app/platform_bundle. The slim image has no xvfb-entrypoint.sh.

Categories

Set in the Apify Console, not in actor.json. Suggested: Jobs, Lead generation, Automation.

Local run

cd scripts/apify-scripts/ats-jobs
mkdir -p storage/key_value_stores/default
cp fixtures/INPUT.json storage/key_value_stores/default/INPUT.json
apify run --purge

--purge clears the default stores only. The named ats-jobs-state store under storage/key_value_stores/ survives, so a second local run with newSinceLastRun: true returns only new jobs. Delete that folder to start fresh.

Unit tests read fixtures only:

$python -m pytest unittests/apify_scripts/test_ats_jobs_actor.py -q

Publish with APIFY_TOKEN (apify push) only after a listing is requested. See scripts/apify-scripts/README.md.