VK Tech Unveils Data Catalog to Keep AI Agents in Check Inside Companies

On 13 October 2026, VK Tech announced the launch of VK Data Catalog, a service that gathers descriptions of corporate data in one place and is meant to make AI agents' work with that data manageable. Little is known so far from the CNews report: it gives no performance figures, no number of supported sources and no pricing. Still, the problem the product targets shows clearly where companies hurt most.

Why agents need a data catalog

A classic data catalog is a metadata layer. It records which tables and data marts exist, what the fields mean, who owns them, where the data came from and who is allowed to see it. For analysts, such a tool has long been routine. With agents, the situation is more acute. A person who sees a column named something like "amt_2" will ask a colleague. An agent will either guess or return a wrong answer with full confidence.

That is why data descriptions are becoming part of the context fed to the model. The more precise the business glossary, and the more clearly it marks which data mart is the authoritative one, the less likely an agent is to calculate revenue from a draft export. The second layer is access rights. An agent acting on behalf of an employee should not see more than the employee can, and every request it makes to the data should leave a trace.

What is known about VK Data Catalog

The announcement is terse. Below is everything the source confirms:

• Developer: VK Tech, the corporate arm of VK.
• Announcement date: 13 October 2026.
• Purpose: managed work by AI agents with corporate data.
• Key function: consolidating data descriptions in a single service (the wording cuts off in the available excerpt).
• Not disclosed: architecture, supported storage systems and DBMS, how agents connect, the licensing model and the price.

It remains unclear how deeply the catalog is built into the agent pipeline. It could be a passive reference that people read and then pass to agents through prompts. It could also be an active layer with a programmatic interface, through which the agent requests context and checks permissions itself. These are fundamentally different products, and the announcement does not say which one this is.

The market and the logic of the release

For VK Tech, the move looks consistent. The company sells infrastructure and data products to enterprise customers, and the customers' next question is obvious: how to let agents near that data without losing control. Western catalog vendors such as Collibra, Alation and Atlan have spent several years promoting metadata as the foundation for AI. In Russia the topic is only gaining weight, and large players with a ready-made ecosystem may claim it faster.

All such products share one weak spot. A catalog is only as valuable as it is complete and current. If the descriptions are outdated or nobody maintains them, the agent will deliver a nicely formatted error. Automating metadata collection therefore matters more than a polished interface.

What to check before a pilot

Companies considering VK Data Catalog should ask the vendor for several specifics:

• List of connectors: which storage systems, data lakes and BI systems the catalog works with out of the box.
• Access method for agents: an API, support for standard protocols such as MCP, or manual use only.
• Permissions model: whether the catalog inherits the sources' access policies and how it records which agent requested what.
• Description automation: who maintains the glossary and how, whether descriptions are generated and how their quality is controlled.
• Deployment: cloud or the customer's own perimeter, which is critical for banks and the public sector.

Until the vendor answers these questions, VK Data Catalog remains a claim on a clear and needed niche, not a proven solution. The coming weeks will show whether documentation and customer case studies appear.