A job board that shows its work.

What Imba Jobs is

Imba Jobs is a job board built for people whose job search has an extra constraint: work authorization. It aggregates openings from companies’ own applicant-tracking boards, classifies each one by level (internship, new grad, experienced), track and country, and reads the posting text for what it says about sponsorship. Every role links to the employer’s own application page. We do not take applications ourselves.

Around the board sit the things an international candidate keeps in other tabs: a crowdsourced OPT tracker, sourced guides to the rules as they stand, hackathons and open-source projects where first pull requests get merged, and a way to connect your own AI agent. Everything works without an account; signing in only syncs across devices.

Imba also makes Imba Interview Tool, a separate desktop app for interview preparation that runs on your own AI provider. It’s an optional subscription ($59.99/month after a 7-day free trial); the job board and everything above stay free.

How sponsorship flags work

Each posting gets one of three labels, and the label is derived mechanically from the posting text — not from a company list, a forum rumor, or what the company did last year. The employer’s track record is shown separately, as described below.

• Sponsors visas. The posting contains a sentence that says the employer will sponsor, supports visa transfer, or is open to candidates who need authorization.
• No sponsorship. The posting contains a sentence that says the employer cannot or will not sponsor, requires authorization without sponsorship “now or in the future,” or requires citizenship or permanent residency (often because of a clearance or export-control rule).
• Unconfirmed. The posting is silent. This is the most common label and it is not a failure state — many routine sponsors never mention it. It means exactly what it says: we do not know.

Technically, the classifier is a set of regular expressions run over the plain text of the posting, with patterns for affirmative language, negative language, negations (“not able to sponsor”) and citizenship or export-control requirements. When a pattern matches we store the sentence it matched and show it to you as evidence next to the flag, so you can judge whether the sentence means what we think it means. A company page rolls the labels up across all of a company’s open roles so you can see whether “no sponsorship” is a one-team template or a policy.

Beside the posting’s label sits a second, separate signal: the employer’s own record in the USCIS H-1B Employer Data Hub, which counts the H-1B petitions USCIS approved and denied for each petitioner name, by fiscal year. We link a company to those records only when a petitioner’s legal name matches exactly under one consistent tax ID, or when we have checked the link by hand; otherwise its page says we couldn’t match it, rather than guessing. When a US posting says nothing about sponsorship and the employer had at least 10 new-employment approvals in the latest full fiscal year, or at least 25 approvals in the hub’s most recent fiscal year to date, the role carries a quieter “H-1B filer” note. That record shows the employer files petitions, not that a particular role is open to sponsorship, and a posting that rules sponsorship out always takes precedence.

Flags are wrong sometimes. Boilerplate gets pasted into the wrong template; a sentence about one location is read as applying to all; a double negative defeats the pattern. Treat a flag as the beginning of your check, read the quoted sentence, and when it matters, verify against the public filing data described in our guide on how to tell whether a company really sponsors. If you spot a wrong flag, tell us — the sentence we quote makes it quick to fix the pattern.

Where the data comes from

• Jobs are fetched directly from the public job boards companies publish on five applicant-tracking systems — Greenhouse, Lever, Ashby, Workable and SmartRecruiters — on a schedule. Companies hiring only through other systems or their own career sites are not covered yet. We store the posting text so we can classify it and show evidence; the apply link always goes to the employer.
• H-1B history comes from USCIS: its yearly H-1B Employer Data Hub files for fiscal years 2019 to 2022, and the live hub, which USCIS updates quarterly, from 2023 on. We refresh it weekly. Counts are workers on petitions USCIS decided, under the petitioner’s own name.
• Hackathons come from Devpost’s public listings.
• Open-source repositories come from the GitHub API, including the outside-contributor merge statistics we show as “merge odds.”
• OPT timelines are submitted by visitors to the tracker, anonymously and unverified. They are not USCIS data and we say so on every page that shows them.
• Guides cite primary sources — regulations, agency pages, court filings and law-firm analyses — inline and in a sources list, and each carries a “last reviewed” date.

What the agent feature does — and does not do

Imba Jobs exposes a documented endpoint that speaks the Model Context Protocol, so an agent you already run — Claude Code in a terminal, a connector in claude.ai, or any other MCP client — can read your profile, your queue, the details of a job and a tailored “apply pack” for it, and can record what it did. You create a key for the agent, label it, and revoke it whenever you like. Keys are stored as hashes; we cannot read them back.

We never submit anything ourselves. Imba Jobs does not run agents, does not hold your passwords for other sites, does not solve CAPTCHAs, and does not press submit on anyone’s application. Your agent acts on your instructions, with your information, within the limits you set — daily caps, a fit threshold, tracks and locations — and the responsibility for what it submits is yours. Our guide on letting an AI agent apply for you explains the platform terms, employer policies and attestations involved, and the limits we recommend.

Where AI is used

Three features can use a language model: parsing an uploaded resume into a profile, checking your fit for a role, and drafting an apply pack. Each has a heuristic version that runs with no model at all, and a model-backed version that runs only when an AI provider is configured for the deployment (here, MiniMax (mainland China platform)). Resume parsing sends the resume text. Fit checks and apply packs send the posting and a summary of your profile — name, location, work-authorization and sponsorship status, experience, education, skills, projects and the roles you are looking for — plus, for an apply pack, the draft letter we wrote. Your email address, phone number, links and standard answers are left out. The privacy policy has the full list and the provider’s terms. Sponsorship flags, job classification and the OPT statistics are not model-based; they are patterns and arithmetic, which is why we can show you exactly why a result came out as it did.

Principles

1. Sponsorship truth over volume. Show the sentence behind every flag. “Unconfirmed” is honest, not a failure.
2. Your agent does the applying. We provide clean data — profile, queue, job, apply pack — and record outcomes.
3. Every page is useful logged out. Sign-in only syncs.
4. Fast, quiet, editorial. No fake urgency, no dark patterns, no padding.

Contact

Corrections, wrong flags, companies to add, data questions, or anything about your data: no contact address has been published for this deployment yet. You can see and delete your own data on your profile page without one. See also our privacy policy and terms.