By Dr. Adi Hod, Founder & CEO, and Yossi Avni, CTO, Velotix
Velotix Atlas, part of Velotix v3, introduces a new approach to enterprise data intelligence, enabling AI to autonomously discover, structure, and optimize knowledge without relying on human-defined schemas or labels. It is a significant technological step for Velotix, and the first in a series we have in development.
The Problem With How Enterprise Data Is Organized Today
Enterprise data is organized according to how people think about it. Data modelers name the entities. Stewards write the business glossaries. Ontology engineers draw the class hierarchies. Even where machine learning assists, the target structure, the label vocabulary, and the measure of success are all chosen by people.
That creates three concrete limitations.
The structure inherits the limits of human attention. A person can review one schema, a sample of rows, and a handful of documents at a time. A large enterprise holds tens of thousands of tables, hundreds of thousands of columns, millions of lines of procedural code, and billions of logged queries. No team can hold that estate in view at once, so the relationships that only appear when everything is compared together are never found.
The structure can only express what someone already thought of. A connection between a behavior recorded in one system and a signal recorded in another does not appear in any schema, because nobody had a reason to write it down. The organization knows it implicitly, in the gaps between its systems, and has no mechanism to capture it.
And the structure was built for the wrong consumer. Schemas and catalogs were designed for human readability, at a time when people were the ones reading the data. Today most enterprise data is consumed by search engines, BI query planners, and AI agents, which need something different.
Existing solutions do not remove this ceiling. GraphRAG, ontology-generation tools, and data catalogs all depend on human-defined schemas, predefined labels, or a limited set of sources. They help AI retrieve more effectively from a structure people designed. The structure itself remains the constraint.
What Our Technology Does Differently
Velotix Atlas takes a fundamentally different approach. Its core innovation combines multiple specialized AI agents, reinforcement learning, and self-generated verification to discover previously unknown relationships and create machine-native knowledge structures optimized for advanced AI reasoning.
Velotix Atlas analyzes an entire enterprise’s data, code, and usage patterns simultaneously, discovering hidden relationships and concepts beyond human-defined structures. It transforms those discoveries into reasoning-ready knowledge, allowing AI models to perform complex, efficient, and verifiable multi-step reasoning.
Rather than teaching AI to understand human-organized data, Velotix Atlas enables AI to invent and continuously optimize data structures for its own reasoning capabilities.
Developed at Agent Speed
Velotix Atlas is currently in rapid research and development, with a strong focus on translating its patented architecture into practical enterprise applications. That research and development is carried out largely by agents.
This matters beyond the novelty. We are building a system for governing agentic access to enterprise knowledge using agents working on our own codebase, under our own controls. Every assumption we make about what an agent needs, what it should be allowed to reach, and what has to be verified before it acts is tested on us before it reaches anyone else. It is also why the governance constraints are part of the architecture rather than added to it later.
How This Strengthens Velotix
Traditionally, Velotix has focused on discovering, classifying, securing, and governing enterprise data. Velotix Atlas represents a major technological evolution. Beyond protecting data, Velotix aims to autonomously understand, restructure, and optimize it for AI reasoning, discovering hidden relationships and creating entirely new knowledge structures beyond human-defined schemas. This positions Velotix to evolve from a Data Security Platform into an AI-native Data Intelligence and Governance Platform.
By autonomously discovering hidden relationships, behavioral patterns, and new data concepts, Velotix Atlas enables richer, more intelligent classification and provides deeper context for access governance.
Governance is preserved throughout. Existing security policies, sensitivity classifications, and masking rules are carried into the newly generated knowledge structures, allowing AI agents to consume enterprise knowledge within a governed environment.
For customers, this opens three capabilities:
Discover hidden knowledge. Identify previously unknown patterns, relationships, and business insights across disconnected enterprise systems.
Enable advanced AI reasoning. Allow AI agents to answer complex, cross-system questions using verified facts and logical relationships, rather than relying solely on conventional RAG.
Automate data intelligence. Generate AI-ready knowledge structures, semantic models, and governed agent tools without extensive manual data modeling.
The fundamental shift is from governing data based on what humans have defined to governing data informed by what AI can autonomously discover and understand. Customers move from protecting and accessing the data they already understand to discovering and using knowledge they did not previously know existed.
The Innovation Ahead
Velotix Atlas is the first step in a series of innovations and patents now in development. Taken together, our aim is for them to make Velotix, within a short time, the most advanced platform in the world for securing, governing, and using enterprise data and knowledge.
The next step is already underway, and it deserves an article of its own. Stay tuned.
See how policy-based access control governs agent-initiated data access – book a demo.
