DataVendor is now in public beta

On October 22, we announced the official launch of the public beta of DataVendor.

We believe that data is the fundamental bottleneck for model progress. DataVendor provides the solution for delivering data at scale, as the definitive marketplace both for buying and selling data. This is the culmination of two years of hard work, building relationships with AI labs to forefront responsible procurement of high-quality data at scale. Prior to this launch, more than 4,000 vendors have already listed over 50,000 assets. Here's how it works.

On DataVendor, vendors set their own prices, and buyers can see evidence of a listing's quality before they pay. Using our platform, you can browse codebases, RL environments, and operational data directly from the companies that made them. You can either use our UI, or easily connect your agents via the API.

Every listing is run through robust quality checks before going live. We always make sure the listing text doesn't contain credentials or personal information, such as emails or phone numbers. Our other required checks are tailored to different listings. For example, codebases must come with a readable copy of the repository, with its Git history and pull requests. Once a listing passes all of our checks, it is awarded our DV quality certified badge. Buyers can access the results behind an individual badge being awarded, and filter our catalog to show only certified listings.

Our parent company, HUD, provides the infrastructure to run QA/QC on tasksets so sellers can identify and fix problems before making a listing. Because DataVendor runs on HUD's platform, a purchased taskset and its environments can be copied straight into the buyer's own team, ready to run.

Before buying, buyers can access metadata, delivery details, price, quality evidence, taskset previews, and a redacted file tree, although never the raw files. Sellers remain protected, while buyers can feel confident that they are receiving data that adds value and meets their precise requirements. For many of our listings, buyers can also buy a sample before committing to a purchase, consisting of one environment and one task. After sampling, buyers can either buy at the listed price, or make an offer.

Buying outright or buying a sample stays anonymous to the vendor, whereas an offer shows the vendor the buyer's organization, so that the vendor can choose whom to accept. Once a purchase is paid for, the buyer retains access even if the vendor later edits or unpublishes the listing.

If you want something that isn't listed, post a request and our network of data vendors will source the highest quality supply to fit your needs, either by listing it on the marketplace or through hand-matching.

Requests from labs often arrive in batches around their model-training runs. Right now, we see the strongest demand for enterprise software codebases and for cybersecurity, legal, and finance data.

We also provide advisory value estimates for private codebases prior to listing. It's easy to use our estimator, which connects to a repo on GitHub, GitLab, or Bitbucket with a single click, and returns a price range in minutes.

A sale on DataVendor is a license for commercial use, so the vendor retains ownership by default. The agreement behind each individual listing sets the terms of the license.

Vendors can sign up at datavendor.ai, and start listing as soon as their organization completes setup. Buyer access is invitation-only. AI labs can request buyer access at datavendor.ai/buy.

Anyone with questions about their data or a buyer's needs can book a call with the DataVendor team.

Frequently asked questions

Our three FAQs answer the most common questions in more detail. The General FAQ covers what DataVendor is, how it makes money, and how it protects both sides, while the Selling FAQ and the Buying FAQ cover each side of the marketplace.