DataVendor by HUD (codebase marketplace) review: who pays, and who it is for

DataVendor is HUD’s marketplace for code and AI training data: vendors list codebases, data bundles and training tasks at their own prices, an AI company that buys a listing pays through DataVendor’s Stripe checkout, and HUD says the seller “receives 80% of the final payout, with 20% covering grading and packaging” (as of October 2026). HUD’s selling guide for companies is written for startups that are shutting down, and centers on their code. HUD here is Human Union Data, Inc., a San Francisco start-up that runs a platform for building AI training environments, not the U.S. Department of Housing and Urban Development.

Can an operating business sell its records on DataVendor?

Not yet, if the records are email, chat or CRM history. DataVendor’s supply types page lists “Operational data (coming soon) — CRM exports, SaaS data, SOPs, process docs, and other records that describe how your company runs,” and says “HubSpot and Salesforce connectors for that category are not live yet” (checked October 5, 2026). What it takes today is RL environments, tasksets, codebases connected from GitHub, and zip bundles of datasets, documents or media.

HUD’s selling guide opens, “Shutting-down startups can sell codebases and internal data to AI labs for real cash”, and asks for each repository “with its full commit history intact,” plus Jira or Linear ticket exports, snapshots of internal docs and a one-page context brief. A HUD go-to-market staff member’s June 2026 post describes the codebase it wants as private, with “multiple contributors, a lot of commits, built over months or years — a dead or pivoted startup/SaaS is the classic case.” An operating company that wants to license its records can compare the providers that say they work with operating companies in the buyers directory.

Who pays the seller, and what does HUD keep?

The buyer pays through DataVendor, and HUD says it keeps 20%. Vendors “set per-asset-type pricing — one unit price per kind (per task, per environment, per repo),” and “Buy now checkout runs through Stripe” (supply types). As of October 2026, HUD describes these payments:

Payment	Who pays	Who receives	What HUD publishes
The listing price	The AI company buying	DataVendor, through Stripe checkout	Set by the vendor per repo, task or environment
The seller’s share	HUD	You	“80% of the final payout” (comparison page, June 24, 2026), paid by “USD wire or ACH on a predictable invoicing schedule” after delivery in verified milestones (HUD guide, July 7, 2026)
HUD’s share	Kept from the buyer’s payment	HUD	The other 20%, “covering grading and packaging”

The split comes from HUD’s own comparison of places to sell a codebase, which recommends DataVendor for modern codebases. The datavendor.ai pages Data Licensing Report read do not state it, and the help page keeps its answer to “How does HUD make money?” behind sign-in, so get the split written into the vendor agreement. On timing, the June 2026 staff post said payment comes within 7–14 days; HUD’s guide gives no number of days.

What will a codebase sell for on DataVendor?

HUD’s own figures run from a $5,000 starting price to $220,000, and none is presented as a completed sale. As of October 2026:

Figure	Where it appears	What it is
“Starts at a $5,000 baseline per codebase and increases with grading”	HUD comparison page, June 24, 2026	HUD’s starting price
“Frontier AI labs pay up to $10k for a qualifying one”	HUD staff post on DEV, June 2026	A call for codebases
$68K to $220K across six codebase cards, from a Terraform library to an eight-year payments monorepo	DataVendor homepage	Cards with no “sold” or “example” label
$50,000 to $300,000 for “typical successful codebase deals”	RL Atlas directory, reviewed September 9, 2026	Figures RL Atlas says are company-reported and “have not been independently verified”

HUD’s comparison page says the final payout “depends on codebase quality, documentation, production history, and clarity of rights, so weaker submissions may remain closer to the baseline.” DataVendor’s free codebase check promises “a price range and demand signal in minutes.” Neither gives the price a comparable listing sold for, and that is the figure to ask HUD for. How much is my company’s data worth? keeps closing-company prices apart from operating-company ones.

Does a buyer get exclusive rights to my code?

Not by default, according to HUD. Its guide says “Each sale grants a non-exclusive license” and “Most labs assume non-exclusive terms unless they negotiate and pay for exclusivity,” and the staff post says “It’s a license, not a sale — you keep ownership,” adding that “the same codebase can be licensed to more than one buyer.” Codebases are bought whole: buyers “purchase the whole listing or a platform-chosen sample,” and samples are offered only for tasks and environments (supply types).

Before buying, a buyer sees a “redacted file tree — folder names are real, file names are placeholders,” and the code is then delivered as a self-hosted download. Folder names can carry customer or project names, so check them before listing. What a buyer may do with the code, and for how long, is not on HUD’s public pages: vendor documentation sits behind sign-in, and HUD’s terms of service, last updated September 22, 2025, cover its website, evaluations platform and SDK, with no provisions for vendors. The terms to settle are listed in Am I selling my data or licensing it?.

Who removes personal data, and who confirms we own the code?

The seller does both. HUD’s guide says “Scrub any PII before delivery. HUD will not accept repos containing unscrubbed personal data,” and “Confirm you actually own what you plan to sell. Your employee IP assignments should be clean.” It also says to “Sign an NDA before any data moves.” The staff post adds that “Sometimes labs need mild processing (e.g. stripping PII) to package the code — HUD gives you tools on the platform to prep and package it easily.”

Commit history records each author’s name and email address, and ticket exports and internal docs often name staff and customers. What does “anonymized” mean? covers what removing personal details involves, and Is it legal to sell company emails and records? lists the documents to check before anything is delivered.

Who is behind HUD, and what is its record?

A Y Combinator start-up founded in 2025. Its YC profile names its founders, Jay Ram (CEO), Lorenss Martinsons (co-CEO) and Parth Patel (CTO), and lists 24 employees; HUD’s terms give its address as Human Union Data, Inc., 2261 Market Street STE 86509, San Francisco. In June 2026, HUD announced a $16 million Series A led by Standard Capital, with Y Combinator and Exceptional Capital among the other investors.

Its record as a marketplace comes from HUD itself. The funding announcement says its “vendor marketplace moves millions of dollars in training data each month,” and the June staff post said HUD had “facilitated a couple million in transactions” in the previous month. DataVendor names no buyers, and Data Licensing Report found no independent coverage of DataVendor as of October 5, 2026. The same HUD comparison page covers Turing’s Project Lazarus and SimpleClosure’s AssetHub, which also deal in closing startups’ code.

What should I ask HUD before listing?

Ask about fees, who pays and when, who can buy, and what a buyer may do with the code:

1. Does the 80/20 split apply to our asset type, and are there any other fees?
2. Which entity pays us, and how long after a purchase?
3. Who can see and buy the listing, and can we exclude particular buyers?
4. What may a buyer do with the code, for how long, and can it pass the code on?
5. Can we read the vendor agreement before connecting a repository?
6. If our company is still operating, when will operational data be accepted, and who would de-identify it?