Company Careers Scraper — Greenhouse Lever Ashby

One input — a list of companies. The Actor auto-detects each company's ATS (Greenhouse, Lever, Ashby or Workable), pulls every live posting through official APIs and returns a single deduped dataset.

How ATS detection works

You pass company slugs — the Actor probes each company on Greenhouse, Lever, Ashby and Workable (one API request per ATS; the fetch itself is the probe). Whichever board answers is fetched in full; companies found nowhere are logged and skipped. Everything is merged and deduped by URL into one dataset. Use atsProviders to skip ATS you know a company isn't on — fewer providers probed per company means fewer upstream requests and a faster run.

Input

Field	Type	Default	Description
companies	array (required)	—	Company slugs (full board URLs also accepted). Slugs are probed on every selected ATS; misses are just logged.
atsProviders	array (multi-select)	all 4 (greenhouse, lever, ashby, workable)	Restrict which ATS get probed per company.
keyword	string	—	Case-insensitive substring match on the job title.
titleExclude	array	—	Drop postings whose title contains any of these substrings (case-insensitive).
locationFilter	string	—	Case-insensitive substring match on the location.
postedSince	integer	—	Keep only postings first published within this many days. Postings with no posting date are dropped when this is set.
remoteOnly	boolean	false	Keep only postings the source ATS flags as remote. Greenhouse-hosted companies never match this filter — see FAQ.
includeDescription	boolean	true	Include a plain-text description snippet per posting.
maxItemsPerCompany	integer	100	Cap postings returned per company (0 = no cap). Each result is a billed event. If the cap actually cuts postings for a company, one extra dataset row flags it (warnings: ["truncated ..."]) — see "Warning rows".
maxItems	integer	200	Hard cap on total postings returned (0 = no cap). Each result is a billed event. If the cap actually cuts postings, one extra dataset row flags it (warnings: ["truncated ..."]) — see "Warning rows".
onlyNewSinceLastRun	boolean	false	Delta/monitoring mode: only output postings not seen on a previous run made with this flag on (see "Delta mode / monitoring").
aiEnrichment	boolean	false	Adds aiKeySkills/aiExperienceLevel/aiWorkArrangement/aiVisaSponsorship per posting via the Anthropic or Mistral API — BYOK (see "AI enrichment").
aiProvider	string	anthropic	Which AI provider runs enrichment: anthropic (default, uses anthropicApiKey) or mistral (uses mistralApiKey).
anthropicApiKey	string (secret)	—	Your Anthropic API key. Only used when aiEnrichment is on and aiProvider is anthropic; billed separately by Anthropic, not by this Actor.
aiModel	string	claude-haiku-4-5-20251001	Claude model for AI enrichment (when aiProvider is anthropic): claude-haiku-4-5-20251001 (fast/cheap) or claude-sonnet-4-5 (higher quality).
mistralApiKey	string (secret)	—	Your Mistral API key. Only used when aiEnrichment is on and aiProvider is mistral; billed separately by Mistral, not by this Actor.
mistralModel	string	mistral-small-latest	Mistral model for AI enrichment (when aiProvider is mistral): mistral-small-latest (default, fast/cheap — matches larger Mistral models on this task), mistral-medium-latest, or mistral-large-latest.
concurrency	integer	8	Companies fetched in parallel (advanced).

What company career pages data does this scraper extract?

One flat JSON record per live posting:

Field	Meaning
ats	Which ATS actually served this posting — "greenhouse", "lever", "ashby" or "workable"
company	Real company display name, resolved per-ATS (falls back to the input slug if unresolvable)
id	Posting ID from the source ATS
title	Job title as posted
department	Department or team where provided
location	Location text (may include remote hints)
url	Direct link to the posting
postedAt	First-published date (YYYY-MM-DD) where provided
employmentType	Employment type where the source ATS provides one — always null for ats: "greenhouse" records, since Greenhouse's API doesn't expose it
remote	true/false/null faithful tri-state — always null for ats: "greenhouse" records (no such signal exists there); genuine true/false for Lever/Ashby/Workable records
snippet	Plain-text description excerpt (optional)
salary	Compensation display text, or null — populated only for ats: "ashby"/"lever" records where the company discloses it; always null for "greenhouse"/"workable" (neither API exposes compensation)
salaryMin / salaryMax	Structured salary range bounds, or null
salaryCurrency	3-letter currency code (e.g. "USD"), or null
salaryPeriod	Pay period the range covers ("year", "hour", etc.), or null
globalId	Stable composite id <ats>:<company-slug>:<id> — unique across the whole dataset even when the same company somehow appears on two ATS
warnings	Array of data-quality notes for this record (e.g. ["postedAt missing"]); empty array when there's nothing to flag
isNew	Only present when onlyNewSinceLastRun is on — always true (already-seen postings are dropped, never emitted with isNew: false)
aiKeySkills	Only present when aiEnrichment is on — array of skills/technologies explicitly named in the posting text, never invented
aiExperienceLevel	Only present when aiEnrichment is on — one of entry/mid/senior/lead/unknown
aiWorkArrangement	Only present when aiEnrichment is on — one of onsite/hybrid/remote/unknown
aiVisaSponsorship	Only present when aiEnrichment is on — true/false only if the posting explicitly states a policy, otherwise null

How to scrape company career pages with this Actor

1. Enter one or more company slugs (stripe, mistral, openai, netguru). Use the company's careers-page slug. When unsure, pass your best guess — slugs are probed on all four ATS providers and misses are just logged.
2. Optionally restrict atsProviders, and set keyword / titleExclude / locationFilter / postedSince / remoteOnly / caps.
3. Run and export JSON, CSV or Excel — or call it over the API:

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("nomad-agent/company-careers-bundle").call(run_input={
    "companies": ["stripe", "mistral", "openai", "netguru"],
    "keyword": "engineer",
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["company"], "|", item["title"], item["url"])

curl -X POST \
  "https://api.apify.com/v2/acts/nomad-agent~company-careers-bundle/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"companies": ["stripe", "mistral", "openai", "netguru"]}'

Output example

{
  "ats": "lever",
  "company": "Mistral AI",
  "id": "7894fd8a-ffc9-4c89-87f0-f8a7b695cf01",
  "title": "Account Executive – AI for Citizens",
  "department": "Business",
  "location": "Paris",
  "url": "https://jobs.lever.co/mistral/7894fd8a-ffc9-4c89-87f0-f8a7b695cf01",
  "postedAt": "2025-07-28",
  "employmentType": "Full-time",
  "remote": false,
  "snippet": "We are hiring an Account Executive...",
  "salary": null,
  "salaryMin": null,
  "salaryMax": null,
  "salaryCurrency": null,
  "salaryPeriod": null,
  "globalId": "lever:mistral:7894fd8a-ffc9-4c89-87f0-f8a7b695cf01",
  "warnings": []
}

A stripe-sourced record in the same run would instead carry "ats": "greenhouse" with "employmentType": null and "remote": null — Greenhouse's API doesn't expose either field. Check ats per record if you need to know which fields to trust. Every record carries the five salary* keys regardless of ats, for a stable merged schema — only Ashby/Lever records ever have them populated, e.g. "salary": "$185K - $325K", "salaryMin": 185000, "salaryMax": 325000, "salaryCurrency": "USD", "salaryPeriod": "year" for an Ashby posting whose company opted in to compensation disclosure.

Delta mode / monitoring

Set onlyNewSinceLastRun: true to turn this Actor into a "what's new" monitor. Postings already seen on a previous run made with this flag on are dropped before push — you are not billed for them, so pairing this with an Apify schedule (cron) means every run only returns, and only charges for, postings that showed up since the last flagged run.

How it works: seen postings are tracked in a dedicated key-value store, keyed by each posting's globalId, capped at roughly 50,000 entries (oldest evicted first). The first run made with the flag on has nothing to compare against yet, so it emits everything — all isNew: true. Every emitted record gets isNew: true stamped on it; there's no isNew: false in the output, since unseen postings just aren't included.

Runs made with the flag off never read or write this cache — turning it on and off between runs is safe and has no side effects on normal runs.

AI enrichment

Turn on aiEnrichment and supply your own anthropicApiKey (or mistralApiKey, with aiProvider: "mistral") to add four AI-extracted fields to every posting. Enrichment runs once on the final merged, deduped set of postings across all probed ATS providers — not once per ATS — so you're never billed extra provider calls just because a company happens to be findable on more than one board.

Field	Meaning
aiKeySkills	Specific skills/technologies/tools explicitly named in the title or description — the model is instructed to never invent one.
aiExperienceLevel	One of entry / mid / senior / lead / unknown.
aiWorkArrangement	One of onsite / hybrid / remote / unknown.
aiVisaSponsorship	true / false only when the posting explicitly states a sponsorship policy, otherwise null.

The extraction prompt is explicit about never guessing: when the text doesn't clearly support a value you get "unknown" / null / an empty array, not a fabricated answer. Pick the model with aiModel (Anthropic: claude-haiku-4-5-20251001 default, fast/cheap, or claude-sonnet-4-5 higher quality) or mistralModel (Mistral: mistral-small-latest default — matches larger Mistral models on this task, mistral-medium-latest, mistral-large-latest).

Postings are batched (~12 per call) through whichever provider's API you picked (aiProvider). Your Anthropic or Mistral API key is billed separately by that provider, not by this Actor. Rough cost with Haiku or Mistral Small: enriching 100 postings runs well under $0.05 in provider token spend (short prompts, small JSON replies); Sonnet/Mistral Large cost roughly 4-5x that for the same batch.

Enrichment needs a posting's description text even when you have includeDescription off — this Actor fetches it internally for enrichment either way, then still honors your includeDescription choice for what actually ends up in the output snippet field.

If aiEnrichment is on but no matching key is available (anthropicApiKey/ANTHROPIC_API_KEY for aiProvider: "anthropic", or mistralApiKey/MISTRAL_API_KEY for aiProvider: "mistral"), enrichment is skipped: you get one extra dataset row explaining why (warnings: ["aiEnrichment skipped: no anthropicApiKey or mistralApiKey provided"]), a run status message, and every other posting is still returned normally, just without the ai* fields.

This is the same class of field fantastic-jobs' career-site-api prices a whole tier on (ai_key_skills, work arrangement, visa signals) — comparable output here, opt-in and BYOK instead of bundled into every row's price.

Warning rows

Alongside normal posting records, this Actor can push a handful of extra rows that are all null except company/ats/warnings — same flat schema as a posting record, so a dataset-only consumer (no access to the run log) still sees every problem:

Trigger	Example warnings
A requested company slug 404'd (or hard-failed) on every ATS probed	["not found on any ATS probed (greenhouse/lever/ashby/workable): 'acme-corp' (404 on every probe)"]
maxItemsPerCompany actually cut postings for one company	["truncated for 'acme-corp': maxItemsPerCompany=100 returned 100 of 137 matching postings — raise maxItemsPerCompany to get the rest"]
maxItems actually cut the total run	["truncated: maxItems=200 returned 200 of 340 matching postings — raise maxItems to get the rest"]
aiEnrichment requested but no API key available	["aiEnrichment skipped: no anthropicApiKey or mistralApiKey provided"]

None of these rows are billed as a result event — only genuine postings are. Pass 50 company slugs in one run and you can tell exactly which ones need fixing straight from the dataset, instead of hunting through the run log.

Integrations

Export results as JSON, CSV or Excel/XLSX, or pipe them straight into Make, Zapier or n8n. Call this Actor synchronously with run-sync-get-dataset-items, or plug it into any AI agent through the Apify MCP server.

Pricing

Pay per event: $0.05 per Actor start and $0.004 per posting returned. 100 postings ≈ $0.45. No subscription — pay only for what you fetch.

If you turn on aiEnrichment, your Anthropic or Mistral key is billed separately by that provider for the enrichment calls themselves — see "AI enrichment" above for a rough per-100-postings cost estimate. Delta mode (onlyNewSinceLastRun) only reduces cost: already-seen postings are dropped before the billed push step.

Use cases

• Watchlists: monitor hiring across target/competitor/portfolio companies
• Recruiting agencies: one feed over every client's ATS
• VC talent teams: portfolio-wide open-roles dashboards
• Job boards that aggregate by company rather than by board

FAQ

Is it legal to scrape company career pages? The data comes from the ATS providers' official, public, unauthenticated JSON APIs — the same data any visitor sees on the company's careers page. Review the providers' terms for your use case.

Do I need an API key or login? No. These are public job-board APIs — no authentication of any kind.

What if a company isn't found? The run continues with the other companies, and — since 2026-07 — you also get one dataset row for it: {"company": "<slug>", "ats": "...", "warnings": ["not found on any ATS probed (...): '<slug>' (404 on every probe)"]} (or "fetch failed on every ATS probed ..." for a network/5xx failure rather than a clean 404). Previously this was only visible in the run log, easy to miss across a 50-slug batch. Full board URLs are also accepted and reduced to slugs automatically. See "Warning rows".

Why do some records have remote: null and employmentType: null? Those are ats: "greenhouse" records. Greenhouse's public API never exposes a remote/workplace-type or employment-type field, so this Actor reports null rather than guessing. Lever/Ashby/Workable records populate both fields whenever the source states them.

How fresh is the data? Every run hits the ATS APIs live. No caching layer in between.

Something broken or missing? Open an issue on the Actor's Issues tab — it is monitored and fixes ship fast.

Related Actors

• Greenhouse Jobs Scraper — Company Job Boards
• Lever Jobs Scraper — Company Job Postings API
• Ashby Jobs Scraper — Company Job Boards API
• Workable Jobs Scraper — Company Job Boards
• Web Developer Jobs Scraper — 10 Boards in One
• LinkedIn Jobs Scraper — No Login, No Cookies