Career Site Jobs Monitor (Greenhouse, Lever, Ashby, Workday +4)

Give a list of companies and get their current job openings straight from their ATS. On scheduled runs you can emit only what is new since the last run.

Supported ATS boards (public JSON/XML, no login, no browser):

ATS	Shorthand example	Board URL example
Greenhouse	greenhouse:stripe	https://boards.greenhouse.io/stripe
Lever	lever:spotify	https://jobs.lever.co/spotify
Ashby	ashby:ramp	https://jobs.ashbyhq.com/ramp
SmartRecruiters	smartrecruiters:BoschGroup	https://jobs.smartrecruiters.com/BoschGroup
Recruitee	recruitee:optics11	https://optics11.recruitee.com
Workable	workable:runware	https://apply.workable.com/runware
Personio	personio:personio	https://personio.jobs.personio.de
Workday	workday:nvidia.wd5.myworkdayjobs.com/NVIDIAExternalCareerSite	https://nvidia.wd5.myworkdayjobs.com/en-US/NVIDIAExternalCareerSite

Input

companies (required, array of strings). Each entry can be:

1. ATS board URL — e.g. https://boards.greenhouse.io/stripe.
2. Shorthand ats:slug — e.g. greenhouse:stripe, lever:spotify, ashby:ramp, smartrecruiters:BoschGroup, recruitee:optics11, workable:runware, personio:personio, workday:nvidia.wd5.myworkdayjobs.com/NVIDIAExternalCareerSite.
3. Any careers-page URL — the ATS is detected by scanning the page HTML for known ATS host names.
4. A bare company name/slug (guess mode) — tried against the major ATS in order until one returns jobs.

Optional filters:

• titleKeywords / excludeTitleKeywords — keep/drop jobs by title keyword.
• locationKeywords — keep jobs whose location matches any keyword.
• remoteOnly — only jobs flagged or named remote.
• postedWithinDays — skip jobs older than N days (when postedAt is known).
• onlyNew — emit only jobs not seen in previous runs (state is stored in the actor's named key-value store).
• includeDescription (default true) — set to false to save time on SmartRecruiters/Workday by skipping per-job detail calls.
• maxItemsPerCompany (default 1000) and maxItems (default 5000) caps.

SmartRecruiters and Workday are fetched in two steps to save money on capped runs: the list of jobs is fetched first, then per-job detail requests are made only for the jobs that will actually be pushed (after filters, onlyNew and the global maxItems budget are applied). A maxItems=1 run therefore makes exactly one detail request, not one per board job.

Output

One dataset item per job:

Field	Type	Notes
company	string	input entry exactly as given
companyName	string|null	from the ATS; falls back to the prettified board slug/tenant when the ATS is silent
ats	string	greenhouse, lever, ashby, smartrecruiters, recruitee, workable, personio, workday
boardSlug	string	board slug used for the API
jobId	string	ATS job id
title	string	job title
department, team	string|null	from the ATS
locations	string[]	one or more locations
remote	boolean|null	explicit flag or heuristic on the location
employmentType	string|null	e.g. Full-time
url, applyUrl	string	canonical posting and apply URLs
postedAt	ISO string|null	publication date (Workday uses an approximation)
postedAtRaw	string|null	raw date text (e.g. Workday "Posted 3 Days Ago")
updatedAt	ISO string|null	last update
salary	object|null	{min,max,currency,interval}
descriptionHtml, descriptionText	string|null	null when includeDescription=false
isNew	boolean	true for items not seen in earlier runs
scrapedAt	ISO string	when the item was collected

A run also writes a SUMMARY key-value record with one status row per company (ok | skipped | not_found | unsupported | blocked | error, plus job counts; skipped = not processed because maxItems or your charge limit was reached). If every company fails, the run fails loudly instead of returning an empty dataset.

What is NOT collected

No personal data: recruiter / hiring-manager / applicant names, emails, avatars, application questions, mailbox_email, open_questions or data_compliance fields are never in the output. Targets are public ATS APIs; no logins, no CAPTCHA bypass, no proxies.

Scheduling & "only new" recipe

1. Run once with onlyNew: false to seed the state.
2. Schedule the actor (e.g. daily) with the SAME companies list and onlyNew: true. Only jobs published after the previous run are emitted.
3. Change the companies list whenever your target set changes — state is keyed on the sorted list, so a new list starts fresh automatically.

State entries not seen for 90 days are pruned automatically.

Limits & behaviour

• Per host: max 2 concurrent requests, >=300 ms between requests, 30 s timeout.
• Retries (up to 3) with exponential backoff on transient errors and 429; a Retry-After header is honoured (capped at 60 s).
• A 403, CAPTCHA or JS-challenge page is recorded as blocked and never retried.
• Items are validated against the output schema; if more than 20% fail, the run fails loudly so a board layout change is never reported as "no jobs".

Development

npm install
npm test        # offline, fixtures only