The USA job
market, unranked
and updated 2x/week.
JobJar polls ten USA-focused sources twice a week and surfaces the result without a recommendation layer. United States jobs only — no international listings. No sponsored slots. No “jobs you might like.” Just the listings, the filters, and the ability to save what matters.
Four steps between a posting going live and you finding it.
Scan 10 feeds
Twice a week (Mon, Wed) we pull from ATS systems, job boards, APIs, and curated community sources.
Tag seniority & remote
Each listing is automatically classified — Entry, Mid, Senior, Principal, Exec — and remote status parsed from the body.
Store & expire
New jobs are written to the index immediately. Anything older than 60 days is dropped so stale posts don't clutter results.
Query, filter, save
Search by title or company. Filter by city, seniority, remote, or source. Exact-match narrows to precise strings.
Every USA listing on JobJar comes from one of these.
Applicant-tracking systems. The raw source — jobs appear here the moment a company opens a req.
USA-focused job boards. Indexed 2x per week, USA jobs only.
Structured data feeds. The cleanest metadata of the ten — Adzuna is queried through its US-only endpoint, location pre-tagged.
Tech-adjacent and startup-heavy. Noisy but catches roles boards miss.
“No ranking. No
sponsored slots.
Just the USA job market.”
Chronological, always
Jobs are ordered by the date posted to the source — not by clicks, applications, or engagement signals. The newest roles surface first, every search.
Every source is named
All 10 feeds are listed publicly on this page. No mystery partners, no undisclosed data suppliers.
Saving is free — and optional
Search and browse without an account. Create one only when you want to save jobs to your Jar and mark them applied.
60 days, then out
Listings older than 60 days are pruned from the index entirely. What remains is actionable — no ghost postings, no filler.
Self-running, self-cleaning, and the AI bill is zero.
Boring on purpose
Next.js on Vercel, Postgres on Neon, crawls on GitHub Actions cron. No servers to babysit — the whole system runs unattended twice a week.
Regex before AI
Deterministic title and location rules decide roughly three of every four listings instantly. Only the genuinely ambiguous remainder is sent to a language model.
Never judge twice
Every verdict is remembered. Rejected URLs are never re-classified, and employers with a pure track record decide their own future listings — the index gets cheaper as it learns.
$20/crawl → $0
Classification originally ran every listing through a paid model. Rules, batched calls, and company memory cut the AI workload far enough that a free-tier model now handles all of it.
Ready to search
the whole market?
// 93,402 listings indexed · updated 06:39 UTC