78 applicant-tracking systems, one schema

freehire’s promise is boring on purpose: every posting, normalized into one shape, deduplicated, linking back to the company’s own careers page. The boring part is the point. The unglamorous part is that “one shape” hides a long tail of applicant-tracking systems that agree on almost nothing about how you’re supposed to read them.

As of this week the catalogue pulls from 78 ATS platforms — 2.9M open postings between them. Here’s what integrating that many taught us, grouped not by vendor but by the kind of problem each one hands you.

The clean ones

A handful of platforms make aggregation genuinely boring — the good kind. Lever (api.lever.co/v0/postings), Ashby (api.ashbyhq.com/posting-api/job-board), Greenhouse, SmartRecruiters: a public, documented JSON endpoint per board, sane pagination, stable field names. You point an adapter at a company’s board token and you’re done. Most of the long tail of company entries in freehire — one line of YAML each — rides on these four, because adding a company is trivial when the platform already speaks clean JSON.

If you’re a job board deciding what to build: this is the bar. A boring, stable JSON feed is the single kindest thing you can do for everyone downstream of you.

The heavyweights

The five biggest by volume tell the story: Workday (831,217 open postings), Oracle (291,963), SmartRecruiters (257,443), Greenhouse (178,084) and iCIMS (122,532). Workday alone is a quarter of everything. At that scale the quirks stop being curiosities and become correctness problems: a paginated endpoint that reports a total of zero on a transient error will, if you trust it blindly, “close” tens of thousands of live jobs in a single bad run. Volume forces you to treat every count the source gives you as a claim to verify, not a fact.

The streamers

Some platforms don’t hand you a paginated API at all — they hand you the whole thing at once. nofluffjobs answers its listing as a single ~60 MB document; several isolved tenants ship their entire job sitemap as one stream. You cannot buffer these into memory and parse them; the adapter has to stream-decode as the bytes arrive. It’s a good reminder that “the API” is sometimes just “a very large file,” and your reader has to be built for that from the start.

The stateful ones

Then there are the platforms where you can’t make the first useful request until you’ve earned a credential. Cornerstone embeds a JSON config blob with a short-lived JWT read token on the careers page; you scrape the token, then send it as a Bearer to the real search endpoint. Taleo wants a session and the right timezone before it will talk. None of this is documented — you infer the handshake by watching what the browser does, then reproduce it.

The hydrating ones

The last group has no API in any honest sense. The jobs live inside the page — embedded JSON, JSON-LD, or a framework’s server-rendered hydration payload (Next.js flight data, an SSR data island). The “integration” is really an extraction: find where the structured data hides in the HTML and pull it out before it becomes pixels.

The takeaway

One schema is worth the tax. From the outside freehire looks like a single feed with faceted filters; underneath, every one of those 78 platforms needed its own answer to the same three questions — how do I list, how do I paginate, how do I read one posting. The value isn’t any single adapter. It’s that once a posting clears the pipeline, nothing downstream — search, dedup, the API, you — has to care which of these worlds it came from.

The whole pipeline and the full per-source breakdown are open source; the Sources table in the README lists every platform and its live count. If that kind of thing is useful to you — or you just like jobs straight from the source — a ⭐ on GitHub helps other people find it.