DataVendor wants to create a marketplace for private data used by AI labs.

DataVendor is launching a marketplace to sell private data, code, and training environments to AI labs.

A company may hold years of source code, professional communications, industry records, or internal procedures without ever considering them commercially valuable assets. Yet that information may interest laboratories training and evaluating artificial intelligence models. DataVendor is introducing a marketplace connecting data owners with buyers looking for specialized datasets, software repositories, and reinforcement learning environments. Opened in public beta after two years of development, the platform allows organizations to register assets, estimate their potential value, and prepare them for sale. Its premise is that some of the data most useful for improving AI systems resides inside private organizations, beyond the datasets that can be freely downloaded from the internet. Data That AI Labs Cannot Simply Download

AI model development has long benefited from large collections of publicly accessible documents, programs, and images. Certain capabilities, however, require much more specialized material. An assistant investigating software incidents may benefit from historical failures and their fixes. A customer support agent needs examples of real business requests. A coding model can learn from private repositories that connect reported problems, tests, and the subsequent code changes.

Those resources are not necessarily available in public datasets. They may belong to businesses, be restricted by confidentiality agreements, or require substantial preparation before they can be used for training. The difficulty therefore involves more than obtaining enough data. Buyers also need usable formats, documented quality, and appropriate rights.

DataVendor aims to organize that supply through a specialized marketplace. Owners can publish commercial offers, while buyers search for assets relevant to their training or evaluation needs. The service targets both organizations holding technical archives and researchers capable of creating training exercises.

The platform makes broad claims about how much of the world's data remains private, but the percentages used in its marketing materials are not accompanied by a verifiable methodology. They should not be interpreted as an established measurement of the available market. Source Code, Raw Data, and Structured Training Tasks

The marketplace extends beyond conventional datasets. DataVendor distinguishes code repositories, raw data, and structured exercise collections known as tasksets.

A private repository may contain an application's current source code alongside its development history, issue reports, tests, and accumulated fixes. For a laboratory developing coding agents, the relationship between a reported problem and its resolution can be more useful than the final code alone.

Raw data may include company documents, technical logs, or specialized industry records. Its value depends on the domain, quality, and permitted uses. DataVendor supports importing these collections as ZIP archives, while software repositories can be brought in through services including GitHub, GitLab, and Bitbucket.

Tasksets serve a different purpose. They contain structured problems, instructions, and evaluation criteria for measuring model performance. A collection might require an AI system to analyze contracts, repair code, or interact with APIs. Buyers acquire exercises with defined outcomes instead of material intended solely for consumption during pretraining. Reinforcement Learning Environments Become Commercial Assets

DataVendor's support for reinforcement learning environments distinguishes it from marketplaces devoted exclusively to files. These interactive environments allow an AI agent to perform actions, observe their consequences, and receive feedback on whether it accomplished a task.

An online checkout simulation might require an agent to complete a purchase under specific constraints. A customer relationship management environment can reproduce CRM workflows, while a spreadsheet simulation evaluates changes made to individual cells. Each environment supplies the necessary tools and a mechanism for assessing results.

These assets are relevant to laboratories working on agents that operate software. Unlike static question-and-answer datasets, an interactive environment allows a model to attempt different actions, make mistakes, and receive feedback about their outcomes.

DataVendor works with the HUD and Harbor ecosystems for preparing these tasks. Its seller documentation explains that supported reinforcement learning environments are sold alongside their associated tasksets rather than as standalone assets. The commercial offering therefore includes both interactive software and the scenarios and scoring criteria that make it useful for training. Example Offers Worth Tens of Thousands of Dollars

DataVendor's website illustrates potential assets with listed prices. Examples include an online checkout simulation featuring 120 scenarios for $80,000, a CRM environment containing 64 workflows for $48,000, and a spreadsheet-agent environment with 150 sheets for $36,000.

Other examples focus on programming and tool use. An IDE refactoring environment is displayed at $95,000 for 85 repositories, while a terminal operations collection appears at $72,000. The marketplace also shows collections dedicated to contract analysis, code review, and document evaluation.

These amounts are displayed prices or commercial examples, not independently verified completed transactions. The public materials examined do not provide sales records, buyer identities, or evidence of the actual prices paid for those examples.

The value of an asset also depends on more than volume. Specialized training exercises may require extensive expert work, while a large software repository may be difficult to use if its dependencies or history are incomplete. Buyers must therefore consider the material's quality, scarcity, delivery requirements, and licensing terms. Estimating the Value of a Private