The questions peopleask before they trusta number about themselves.

Yes. The full product is free during early access. No card, no trial countdown, and no feature held back to force an upgrade. Paid Pro and Campus plans are planned, and you will see exactly what they add well before anything about your account changes.

A resume and one target role. That is enough for parsing, ATS checks, a completeness read, and a ranked shortlist. Connectors add resolution to the picture, but nothing waits on them.

Create an account, add a resume and the roles you are chasing, then connect GitHub or install the browser extension if you want richer signals. Candidate strength and your first matches follow from that.

The US, India, and remote-friendly UK/EU roles at launch: postings pulled from official ATS boards, aggregators, startup listings, and campus and internship sources, with company records built alongside them. Sources and regions sit behind interfaces, so widening coverage is configuration rather than a rebuild.

The same role posted to four boards collapses into one entry with its sources listed. Listings that fail vetting, meaning no description, no apply route, or recruiter spam markers, are quarantined out of every feed rather than left to waste your evening.

Drop in the URL. It gets parsed into the same structured shape as everything else, scored against your profile, and tracked on the same timeline. Nothing is second class because it arrived by hand.

It decomposes into ten drivers: skill coverage, missing must-haves, seniority fit, domain alignment, compensation fit, location and remote fit, employer signal, your strength against the role bar, posting freshness, and competition. Each driver has a visible weight and a visible value, so you can see which one is carrying the score and which one is dragging it down. It is feature math you can inspect, not a percentage you have to take on faith.

No. Models are used for parsing, enrichment, and drafting, where a bad answer can be caught on review. Ranking runs on deterministic features with versioned weights, which is why the same profile against the same posting produces the same score every time.

Because it is reproducible. Twenty-four fixed checks across parseability, structure, and content, each weighted by severity, computed with zero model tokens. Language suggestions are listed separately and never move the number.

That is the intended reaction, and it is why the drivers are exposed. Open the match, find the driver you disagree with, and either fix the input behind it or discount it deliberately. A score you cannot interrogate is a score you cannot use.

Resumes, code evidence, professional profile data, postings, company records, contacts, and replies detected by the browser extension become structured signals that feed scoring. GitHub connects only with your permission and only with read-only scopes, tokens are stored encrypted, and every connector can be revoked without losing the work built on top of it.

On Greenhouse, Lever, and Ashby, Atlas can submit directly through the official API once you allow it. Everywhere else the browser extension fills the form in your own session, highlights anything it could not map, and you click Submit. Atlas never submits silently, and monthly caps apply.

Applications and outcomes are logged so ranking improves over time, but learning happens on structured signals rather than on your private context. Nothing you wrote is surfaced to anyone else. You can export or delete your data whenever you want.

Yes. Every automated step has a hand-operated version next to it. Atlas is designed to remove the tab management and the tracking spreadsheet, not your judgement about where to spend an evening.