ORBITORBIT searches about 73 million research datasets through the DOI registry that Zenodo, figshare, Dryad, PANGAEA and most institutional archives all mint their identifiers through, and machine-learning datasets on Hugging Face. Every result says whether you are allowed to reuse it, links out to the repository by DOI, and can be exported to a reference manager. A dataset card shows who has used it, and a second tab finds where to deposit your own data. The whole page is free and takes no credit.
This is the distinction the tool exists for. A dataset search that told you only where the files are would be answering the easy half of the question. ORBIT shows every result in one of three states, and the third is the one that matters most.
Openly licensed means the record states a licence permitting reuse — CC0, CC BY, MIT and the like. Licence restricted means a licence is stated and carries limits worth reading, such as non-commercial. Licence not stated means the record names no licence at all, which is common and is not permission: in most jurisdictions the default is all rights reserved. ORBIT will not present a blank field as a green light.
The DOI registry research repositories mint through, read via its own API. That is why one search reaches Zenodo, figshare, Dryad, PANGAEA, Dataverse and a long tail of institutional archives at once, with no arrangement needed with any of them.
ORBIT does not crawl repository pages, and it does not host, mirror or redistribute anybody's data. Every result links out to the source by DOI — the identifier that still resolves after a repository reorganises its site, which for a dataset citation is the whole point.
Repositories mint a separate DOI for every version of a dataset and index the parent record beside them, so raw results repeat themselves — nine of twenty-five rows in a live measurement. ORBIT collapses versions into one row and tells you how many exist, grouping them by the identifiers repositories publish rather than by matching titles, because two different datasets can share a name.
Metadata quality varies enormously. Some records carry a full abstract, named creators and a licence; some carry a title and nothing else. ORBIT marks the sparse ones and names what is missing rather than presenting every hit as equally trustworthy. A thin record can describe excellent data — you simply cannot tell, and you should know that before you spend an afternoon on it.
Every result opens a card. It puts the licence in plain words, lists the versions, size and formats, and gives the citation, view and download counts with their source and date. It names the paper the dataset came from and lists the papers that cite it, newest first, which open as a list in Literature Discovery.
Views and downloads are reported by each repository, and some report none, so the card says when a repository sends no usage figures rather than showing a zero. A short citation list means few recorded citations, not that nobody used the data.
Pick your field and ORBIT lists the repositories that accept data in it, from a registry of about 3,500 research data repositories. Those open for deposits come first, then those that give a DOI, then those that state a certificate. Each says whether it charges a fee and who may deposit, and you can limit the list to Indian repositories.
Zenodo, figshare, Dryad, the Open Science Framework, Dataverse and Mendeley Data take data from any field and are always shown.
There is no catch and no paid tier behind it. Searching, filtering, following the links and exporting are all free and take no credit, and unlike the literature tool there is no paid analysis pass on this page at all. The registry ORBIT reads is free to query, so there is nothing here worth charging you to recover.
Not necessarily, and this is the single most important thing on this page. Being able to download data is not the same as having permission to reuse it. ORBIT shows every result in one of three states: openly licensed, licence restricted, or licence not stated. That third state is common, and it does not mean the data is free to use — it means nobody has said. In most jurisdictions the default is all rights reserved, so a dataset with no stated licence is one to ask the repository about before you build on it.
The DOI registry that research repositories mint their identifiers through, which is why one search covers Zenodo, figshare, Dryad, PANGAEA, Dataverse and a great many institutional archives at once, without ORBIT needing an arrangement with any of them. About 73 million dataset DOIs. Read through the registry's own API — ORBIT does not crawl repository pages, and it does not host, mirror or redistribute anybody's data. Every result links out to the source.
Open its dataset card. It lists the papers that cite the dataset's DOI, newest first, with how many there are in total, and opens them as a list in Literature Discovery. A button asks a second open scholarly-links service for more, including mentions the first does not record. Many papers use data without citing its DOI, so a short list means few recorded citations, not that nobody used it.
Pick your field under "Where to deposit mine" and ORBIT lists the repositories that take data in it, from a registry of about 3,500. Each row says whether it gives a DOI (what a journal's data availability statement needs), whether it states a certificate, whether it charges a fee, and who may deposit. You can limit it to Indian repositories. General-purpose ones like Zenodo, figshare and Dryad are always shown, because they take data from any field.
Yes. Switch the source to Hugging Face. It matches dataset names rather than descriptions, so one or two words work best, and results come most downloaded first. Each shows its licence as tagged, its downloads over the last 30 days, and whether it is gated behind terms you must accept on Hugging Face. ORBIT links to the dataset and never copies it.
Because it is, and hiding that would be dishonest in the direction that costs you time. Metadata quality varies enormously between repositories: some records carry a full abstract, named creators and a licence, and some carry a title and nothing else. ORBIT marks the sparse ones and says exactly what is missing — no description, no named creator, no year — rather than presenting every hit as equally trustworthy. A thinly described dataset can still be excellent; you just cannot tell from the record, and you should know that before you spend an afternoon on it.
Yes. Repositories mint a separate DOI for every version of a dataset, and both the versions and the parent record are indexed, so raw results repeat themselves — in one live measurement, nine of twenty-five rows were duplicates. ORBIT collapses them into a single row and tells you how many versions exist, so you can still go and find an older one if a paper cited it. It groups them using the identifiers the repositories publish, never by matching titles, because two genuinely different datasets can share a name and merging those would hide data from you.
Yes, in four formats and all free: RIS, BibTeX, CSV and a plain Markdown list. The RIS and BibTeX exports mark each entry as a dataset rather than as a journal article — the wrong type here is not cosmetic, because it silently produces a wrong citation in your bibliography. The exports also carry the version and the licence state, so what you read back in six months says the same thing the screen said today.
No. Results are cached so that two people searching the same topic share one answer, but the cache holds no user id and cannot be traced back to anyone — the key is a hash of the search itself. There is deliberately no search history. What a researcher is looking for is a map of work they have not published yet. If you press Save on a paper, that dataset (not the search) is kept in your Library until you remove it.