jobfeeds 0.1.0

One typed, async Python interface to six public job boards: Remotive, Arbeitnow, Greenhouse, RemoteOK, We Work Remotely, and Hacker News Who is hiring

📡 jobfeeds

One typed, async Python interface to six public job boards.

pip install jobfeeds

---

Fetching job postings should not require writing six API clients. jobfeeds gives you Remotive, Arbeitnow, Greenhouse boards, RemoteOK, We Work Remotely, and the Hacker News "Who is hiring" thread through one interface, normalized into one pydantic model, with failures isolated per source.

Extracted from, and used in production by, JobRadar, a live job-market intelligence dashboard.

Quickstart

import asyncio
from jobfeeds import Remotive, Greenhouse, fetch_all

async def main():
    # One source
    jobs = await Remotive().fetch()               # list[Posting]
    print(jobs[0].company, jobs[0].title, jobs[0].url)

    # Configured source
    jobs = await Greenhouse(boards=["stripe", "anthropic"]).fetch()

    # Everything, concurrently, failure-isolated
    results = await fetch_all()                   # list[SourceResult]
    for r in results:
        print(r.source, len(r.postings) if r.ok else f"FAILED: {r.error}")

asyncio.run(main())

The model

Every source normalizes into the same Posting:

Field	Notes
source	adapter name ("remotive", "greenhouse", ...)
external_id	the source's own id
company, title, url	always present
location_raw	as the source states it, unparsed
posted_at	datetime or None, when the source provides one
raw	source-specific payload; descriptions truncated to 8,000 chars
.hash	cross-source dedupe key (normalized company + title + location)

Boards syndicate each other's postings; posting_hash() collapses the duplicates without trusting any board-specific id.

Sources and their quirks

Adapter	Config	Worth knowing
Remotive()		Curated remote listings, JSON API
Arbeitnow(pages=3)	page count	DACH-market coverage; paginates explicitly because the API's own links.next can point at filtered views
Greenhouse(boards=[...], per_board_limit=40, title_pattern=None)	board tokens, cap, optional title regex	Direct-employer postings (Stripe, Anthropic, GitLab, ...). No global search endpoint exists, you choose boards. A default starter set is included
RemoteOK()		Their terms require a visible link back to remoteok.com in what you build. The API's first element is a legal notice, already skipped
WeWorkRemotely(categories=[...])	category slugs	RSS per category; defaults to programming, devops, design, product
HNWhoIsHiring()		Finds the latest monthly thread via Algolia, parses only structured "Company | Role | Location" comments; freeform replies are ignored

All adapters accept timeout= and user_agent=. Public APIs change; every adapter parses defensively and skips malformed entries instead of raising.

Failure isolation

fetch_all() never lets one dead API take down the run:

results = await fetch_all([Remotive(), Greenhouse(boards=["stripe"])])
postings = [p for r in results if r.ok for p in r.postings]
errors = {r.source: r.error for r in results if not r.ok}

Your own adapter joins the party by satisfying the SourceAdapter protocol: anything with a name and an async fetch() -> list[Posting].

Design notes

• httpx + pydantic only. No framework, no ORM, no surprises in your dependency tree.
• Fully typed, ships py.typed.
• No product logic. Filtering, enrichment, and storage belong to your application; this package only does feed access done well. (JobRadar's tech-role filter, LLM enrichment, and database live in its repo, on top of this package.)
• Tests never touch the network. Every adapter is tested against mocked responses (respx) in CI on Python 3.11, 3.12, and 3.13.

License

MIT. Built by Nandor Nagy as part of a public portfolio; the ingestion layer of jobradar.nagysolution.com.