🇩🇪 Arbeitsagentur Jobs Scraper — the German Federal Jobbörse, as structured rows

Turn arbeitsagentur.de/jobsuche — the German Federal Employment Agency's Jobbörse, the single largest vacancy corpus in the German-speaking market — into clean JSON/CSV/Excel rows: employer, PLZ and city, occupation, contract type, working-time pattern, start date, pay where stated, and the Zeitarbeit (Arbeitnehmerüberlassung) classification that tells you whether a posting comes from a direct employer or from a temp-staffing agency.

No login. No API key. No CAPTCHA. Datacenter proxies are enough.

---

📊 What you are scraping (measured live 2026-09-04)

Offer type (angebotsart)	Live vacancies
Arbeit — regular jobs (1)	746,533
Ausbildung & duales Studium (4)	170,391
Praktikum & Trainee (34)	14,516
Selbständigkeit (2)	3,226
Total	~934,700

Churn is roughly 9% a day — on a sample occupation, 276 of 3,109 postings were less than 24 hours old. Set Published within → Today only and this actor becomes a daily new-vacancy feed.

---

🚀 What does Arbeitsagentur Jobs Scraper do?

• 🔎 Searches the way the site does. Keyword (was), location + radius (wo + umkreis), offer type, publication window, working time, contract type, Berufsfeld, Beruf, employer name, private-placement opt-in, disability-friendly, sort order — every filter is wired to real behaviour and every one was verified to genuinely narrow the result set, not just relabel it.
• 🏢 Classifies Zeitarbeit per job. isTemporaryStaffing / istArbeitnehmerUeberlassung is the qualifier a staffing firm, an HR-SaaS SDR or a Personalberatung actually buys on: is this posting a competitor agency, or a direct employer you can sell to? No rival in this lane exposes it.
• 🧩 Breaks the site's 10,000-row ceiling. arbeitsagentur.de refuses to page past 10,000 rows for any single query (page 400 is the last one that returns data — an Elasticsearch max_result_window). Auto-split cuts a bigger query by federal state → publication window → Berufsfeld → Beruf → contract type → working time, then merges and de-duplicates. That is how "all 746,533 German vacancies" is actually deliverable.
• 🧾 Never double-bills. Rows are de-duplicated on referenznummer across every search term, location and auto-split slice before anything is charged.
• 🌍 Emits English aliases. jobTitle, employer, city, postalCode, publishedAt, isTemporaryStaffing … alongside the original German keys, so rows drop straight into a CRM.
• 🔗 Paste any Jobbörse URL. Build the search in the site's own UI, paste the URL, and every query parameter is honoured verbatim — including filters this schema does not expose. A /jobsuche/jobdetail/<referenznummer> link is fetched as a single job.
• 📝 Optional detail enrichment adds the full job-ad text (median 2,386 characters), the employer's street address, the Zeitarbeit and private-placement flags, the required education level, and the partner job board that syndicated the posting.

---

📦 Output — measured fill rates

List fields, measured on a 350-row national sample across 7 varied searches (Pflegefachkraft · Softwareentwickler · LKW-Fahrer · Elektroniker · Buchhalter · Erzieher · whole corpus), 350 of 350 unique:

Field (German → English alias)	Fill
referenznummer → jobId (stable id, dedupe key)	100%
stellenangebotsTitel → jobTitle	100%
firma → employer	100%
ort → city · land · region · breite/laenge → latitude/longitude	100%
plz → postalCode	97.4%
hauptberuf → occupation · alleBerufe	100%
stellenangebotsart → offerType	100%
veroeffentlichtAm · datumErsteVeroeffentlichung · aenderungsdatum	100%
eintrittsdatum → startDate	100%
vertragsdauer → contractDuration	100%
verguetungsangabe → salaryBasis	100%
arbeitszeitVollzeit → isFullTime	95.7%
quereinstiegGeeignet → isCareerChangerFriendly	84.9%
istGeringfuegigeBeschaeftigung → isMiniJob	84.6%
homeofficemoeglich → isHomeOfficePossible	58.9%
arbeitgeberKundennummerHash → employerKey (groups all postings by one company)	51.7%
externeURL → externalUrl (partner-board deep link)	48.3%
strasse → street	41.4%
artDerVerguetung → salaryKind	36.9%
gehaltsspanneVon/Bis → salaryFrom/salaryTo	21.1%
festgehalt → salaryFixed	15.7%
chiffrenummer (anonymised listing)	14.6%

Detail fields (fetchJobDetails: true), measured on 60 enriched rows:

Field	Fill
stellenangebotsBeschreibung → description (394–4,079 chars, median 2,386)	100%
allianzpartnerName / allianzpartnerUrl → partnerJobBoard	100%
istBetreut	100%
istArbeitnehmerUeberlassung → isTemporaryStaffing	85.0%
istPrivateArbeitsvermittlung → isPrivatePlacement	76.7%
istBehinderungGefordert	75.0%
strasse → street	66.7%
hausnummer → houseNumber	48.3%
geforderterBildungsabschluss → requiredEducation	0–50%, occupation-dependent (see limits)

ausbildungsverguetungJahr1/2/3 → apprenticePayYear1/2/3 appear only on Ausbildung postings (measured 4 of 40 on angebotsart=4).

Turn on Include the raw source objects to keep the untouched ergebnisliste / jobdetail payload under raw for any field this actor does not map.

---

💵 Pricing

Pay per event. No monthly rental, no platform-usage surcharge, no charge for starting a run.

Event	Price	When it fires
Per job returned (job-scraped)	$0.0008 / job — $0.80 per 1,000	Each unique job delivered to your dataset
Per detail fetch (job-detail-enriched)	$0.0012 / job — $1.20 per 1,000	Only when Fetch job details is on, and only for enriched rows that actually shipped

Rows ship through Apify's budget-aware push, so delivered always equals billed: if you set a max-total-charge cap, the run stops at the cap instead of handing you rows you were never charged for. Duplicates across overlapping searches are removed before billing.

A full sweep of the 746,533-vacancy Arbeit corpus costs ≈ $597, refreshed as often as you like. A daily "everything posted in the last 24 hours" feed is ~9% of that per day.

---

⚙️ Example inputs

Daily new-vacancy feed for a staffing firm — direct employers only, Berlin + 50 km:

{
  "searchTerms": ["Pflegefachkraft", "Altenpfleger"],
  "locations": ["Berlin"],
  "radiusKm": 50,
  "publishedWithinDays": "1",
  "temporaryStaffing": "exclude",
  "fetchJobDetails": true,
  "maxItems": 2000
}

Map the competition — every Zeitarbeit posting in Bayern:

{
  "locations": ["Bayern (Bundesland)"],
  "radiusKm": 0,
  "temporaryStaffing": "only",
  "fetchJobDetails": true,
  "maxItems": 1000
}

Whole-country apprenticeship sweep past the 10,000-row ceiling:

{
  "offerTypes": ["4"],
  "autoSlice": true,
  "maxItems": 0
}

Paste a search you built on the site:

{
  "startUrls": [
    { "url": "https://www.arbeitsagentur.de/jobsuche/suche?angebotsart=1&was=Elektroniker&wo=K%C3%B6ln&umkreis=25&veroeffentlichtseit=7" }
  ],
  "maxItems": 500
}

---

🧠 How it works

The Jobbörse is a server-side-rendered Angular app: every search page and every job-detail page embeds its complete state as JSON in a <script id="ng-state"> tag. This actor reads that blob — no DOM scraping, no brittle CSS selectors, and no API key.

• Search: GET /jobsuche/suche?<filters> → #ng-state → suchergebnis.ergebnisliste[] (25 rows per page), with suchergebnis.maxErgebnisse as the structurally-read hit count.
• Detail: GET /jobsuche/jobdetail/<referenznummer> → #ng-state → jobdetail.
• The documented rest.arbeitsagentur.de/jobboerse v4 API returns 403 without an app key; this actor never touches it, because the SSR blob is the identical payload with no key at all.

Transport: plain HTTP over Apify datacenter proxies, HTTP/2, de-DE accept-language, responses decoded explicitly as UTF-8 (the default decoding mangles umlauts — "Universität" becomes "Universit?t"). 512 MB, no browser. Verified over 200+ calls with no CAPTCHA, no Cloudflare, no DataDome, no rate limiting. RESIDENTIAL + country DE also works and is exposed as a fallback.

---

⚠️ Honest limits — read these before you buy

1. The 10,000-row ceiling is the site's, not ours. Any single query stops at 10,000 rows: page 400 returns 25 rows, page 401 returns an empty page state. Auto-split works around it by cutting a big query into smaller ones, but if you pin every axis yourself (a keyword AND a city AND a date window AND a contract type…) and the remainder is still over 10,000, the actor logs a warning and returns the first 10,000 for that query. Narrow it further to reach the rest.
2. About 5% of responses come back HTTP 200 with an empty page state. These are retried, never reported as "no results" — a naive parser reads them as zero rows and finishes SUCCEEDED with nothing. Measured: 19/20 clean on the first attempt, 18/18 clean with one retry. Keep Max retries per request at 3 or higher.
3. Page size is fixed at 25. The site ignores size=. Budget one request per 25 rows.
4. geforderterBildungsabschluss is patchy — 10 of 20 across mixed offer types, but 0 of 60 on a nursing/retail/mechatronics sample. Do not build a workflow that depends on it.
5. Salary is mostly absent. gehaltsspanneVon is present on ~21% of rows and festgehalt on ~16%. German postings usually state a collective-agreement reference (verguetungsangabe, 100%) instead of a number.
6. Street address is a detail-page field and present on ~2 of 3 enriched rows — 41% at list level. If you need it on every row, turn on Fetch job details.
7. employerName is exact-match, not fuzzy. Vivantes Netzwerk für Gesundheit GmbH returns ~320 rows; Charité returns 0.
8. Six federal-state names collide with same-named towns. wo=Hessen returns a town in Sachsen-Anhalt. Use the disambiguating form — Hessen (Bundesland) — when you mean the state. The 16 states in that form sum to 98.3% of the national Arbeit corpus.
9. publishedWithinDays only honours 0, 1, 7, 14 and 28. Any other number is silently ignored by the site and returns the unfiltered set; this actor warns and drops the filter rather than pretending it applied.
10. Zeitarbeit "only" costs roughly twice the requests. The site has no positive Zeitarbeit filter — only a negative one — so "only" crawls both sets and takes the difference. Turn on Fetch job details with it to have every returned row verified against the official flag.
11. Some postings are outside Germany. The corpus includes Austrian and other cross-border vacancies. The land field tells you which; filter on it if you only want DE.

---

❓ FAQ

Is scraping arbeitsagentur.de allowed? The site's robots.txt is fully open, verbatim: User-agent: * / Disallow: / Allow: / — with no crawl-delay. This actor reads only public, unauthenticated pages that the Bundesagentur für Arbeit publishes for exactly this purpose. You are responsible for complying with the site's terms of use and with GDPR for any personal data the postings contain.

Do I need an API key or a login? No. Neither. The documented v4 API needs an app key; this actor does not use it.

How many jobs can I get? All of them. ~934,700 across the four offer types. Auto-split handles the site's per-query ceiling.

How fresh is the data? Live — every run reads the site at that moment. Roughly 9% of the corpus turns over daily.

Can I get only new postings since yesterday? Yes. Set Published within to Today only or Last 1 day.

Can I tell a temp agency from a direct employer? Yes — that is this actor's wedge. Use Zeitarbeit → Exclude for direct employers only, or Only for the agencies themselves. Add Fetch job details to have every row carry the official istArbeitnehmerUeberlassung flag. Verified end-to-end: an "exclude" run returned 0/20 agency postings, an "only" run returned 20/20.

Can I group every vacancy from one company? Yes — arbeitgeberKundennummerHash / employerKey is a stable per-employer key (present on ~52% of rows). employerName also filters by exact registered name.

Does it get the full job-ad text? Yes, with Fetch job details on: 100% fill, median 2,386 characters.

Are the German umlauts correct? Yes. Responses are decoded explicitly as UTF-8 — "Fachverkäuferin", "Backstube Wünsche", "München".

How do I search a specific city, PLZ or federal state? locations accepts a city, a postal code, a federal state or Deutschland. Use the <Name> (Bundesland) form for the six state names that collide with towns.

What happens if my run hits its charge cap? It stops cleanly and tells you so. Every row you received was billed and every row billed was delivered — the cap never leaks unpaid rows in either direction.

What happens if there are genuinely no matches? The run exits cleanly with 0 rows and a status message explaining why (no matches vs. blocked), and charges nothing. It does not fail.

Can I run this on a schedule? Yes — Apify Schedules. A "last 1 day, my region, direct employers only" run is the standard setup.

---

⚖️ Legal & fair use

This actor collects public, unauthenticated job postings published by the Bundesagentur für Arbeit, a German federal agency, on a site whose robots.txt explicitly permits crawling of all paths. It sends no credentials, solves no challenges, and forges no signatures.

Job postings can contain personal data (named contacts). How you store, process and use that data is your responsibility, including under the GDPR/DSGVO and the site's own terms of use. Concurrency is deliberately conservative by default (5) because this is a public service funded by German taxpayers — please leave it there unless you have a reason not to.

Not affiliated with, endorsed by, or connected to the Bundesagentur für Arbeit.