--- title: "Atlan vs Illumex: Business Context AI Compared" url: "https://atlan.com/know/ai-agent/semantic-layer/atlan-vs-illumex/" description: "Illumex built Business Context AI, then NVIDIA bought it in 2026. See what that leaves for buyers, and how Atlan's governed context layer compares today." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-09-10" updated: "2026-09-10T00:00:00.000Z" --- --- Illumex called its product Business Context AI: a Generative Semantic Fabric that read metadata and inferred what a table or column meant to the business, without ever touching the underlying rows.[1] NVIDIA acquired the company in February 2026, and Illumex no longer sells that product on its own.[2] Atlan's [Enterprise Data Graph](https://atlan.com/know/enterprise-data-graph/) does that same generation work, but treats it as one of three governed substrates rather than the whole platform, alongside [trusted AI-ready data](https://atlan.com/know/data-catalog-for-ai/) and the skills and procedures a team actually follows. Gartner frames why the generation step matters at all: organizations that prioritize semantics in AI-ready data could raise agentic AI accuracy by up to 80% and cut costs by up to 60% by 2027.[3] That leaves a different comparison than the query implies. There is no live purchase decision between two active vendors. There is a scope question: does a semantics-only tool cover what an enterprise running agents across many systems needs, or does meaning need to sit inside something bigger that also governs and delivers it. | Dimension | Illumex (Business Context AI) | Atlan (context layer) | |---|---|---| | What it was | A metadata-only semantic layer that auto-generated business definitions | A governed context layer covering semantics, data, and procedures | | Current status | Acquired by NVIDIA, February 2026; not sold as a standalone product | Independently operated; sold and deployed as Atlan | | Core mechanism | Generative Semantic Fabric, reading schema and usage metadata only | Enterprise Data Graph, connecting glossary, lineage, ownership, and policy | | Delivery to agents | Tied to NVIDIA's own AI stack, including a NeMo integration | MCP server, model-agnostic by design | | Scope | Semantics and ontology, one substrate | Semantics, trusted data, and skills and procedures, three substrates | | Governance after generation | Not documented as a separate function | Certification, ownership, and access policy travel with the definition | --- ## Atlan vs Illumex: what's the difference? The honest framing starts with what changed. Illumex spent five years building a real, working answer to one question: what does this column mean, generated automatically rather than documented by hand. [Business context for AI](https://atlan.com/know/business-context-for-ai/) was a real category, and Illumex earned genuine credit in it, including a reported $60 million to $75 million acquisition price.[4] Atlan's [context layer](https://atlan.com/know/what-is-context-layer/) answers a wider question: once meaning is generated, who certifies it, does it stay current, and can any agent on any model retrieve it the same way. Both companies took the same architectural bet. Illumex read schema and usage metadata rather than moving the underlying rows, a privacy-preserving pitch. Atlan's [Enterprise Data Graph](https://atlan.com/know/what-is-the-enterprise-context-layer/) works the same way: it maps an ontology to physical assets, reverse-engineered from lineage and usage, without duplicating source data into a second system. Neither vendor asked customers to move data just to generate meaning about it, one reason the two names get searched together at all. Where they diverge is what happens after generation. Illumex's [own AWS Marketplace listing](https://aws.amazon.com/marketplace/pp/prodview-7iity5oddyx36) described a metric store, data dictionary, catalog, and business glossary bundled together, a real product with real capability. It said nothing about certification workflows or freshness checks travelling with a definition once agents from more than one team asked for it. That gap is exactly where [enterprise context silos](https://atlan.com/know/enterprise-context-silos-ai-teams/) form, and it is the gap [self-service analytics governance](https://atlan.com/know/ai-agent/ai-agent-governance/self-service-analytics-governance-build-vs-buy/) treats as its own decision, separate from which tool generates the definition. --- ## What is Illumex? Illumex was founded in 2021 in Tel Aviv by CEO Inna Tokarev Sela, and it built a product it called a Generative Semantic Fabric: software that inferred business-meaning definitions from schema, column names, and usage patterns, rather than the row-level data underneath.[1] It raised a $13 million seed round in June 2024 and named Teva Pharmaceutical among its enterprise users. NVIDIA acquired Illumex in February 2026, folding its team into NVIDIA's Israeli R&D organization; the deal extended a NeMo integration the two companies had already announced in March 2025, rather than starting a relationship cold.[2] Illumex no longer operates as an independent company, and its own domain no longer resolves. The full timeline and deal terms live on [Atlan's Illumex explainer](https://atlan.com/know/ai-agent/illumex-business-context-ai/); this page picks up from there rather than repeating it. What Illumex built worked, on its own terms. An unnamed retailer's ERP migration, spanning 115,000 tables and 1.5 million data columns, reportedly went from a projected two-year manual review to one week of automated workflows, a vendor-reported figure.[1] That is a credible demonstration that automated mapping compresses a genuinely painful process. It says nothing about what happened to those definitions later, or who was responsible once a source schema shifted, the seam [semantic layer vs data catalog](https://atlan.com/know/ai-agent/semantic-layer/semantic-layer-vs-data-catalog/) treats as two separate jobs. ### Core components of Illumex - **Generative Semantic Fabric:** the core engine, inferring meaning from schema and usage metadata rather than raw rows - **Metric store and data dictionary:** turnkey, packaged definitions sold as part of the platform - **Business glossary and catalog:** unified discovery of business semantics, per its own marketplace listing - **NVIDIA NeMo integration:** announced March 2025, months before the acquisition made the tie permanent - **Current status:** acquired by NVIDIA in February 2026; not sold as a standalone product today --- ## What is Atlan? Atlan is the context layer for AI: a governed system that unifies trusted, AI-ready data, business semantics, and the skills and procedures teams already follow into one place every agent can reach. Its Enterprise Data Graph connects glossary terms, ownership, certification state, and [column-level lineage](https://atlan.com/know/ai-agent/data-for-ai/data-lineage-for-ai/) into a single graph over whatever holds the raw data underneath. Where Illumex generated a definition and packaged it into its own product, Atlan generates the same kind of definition, then asks the harder question: is it still correct, who owns it, and can it be trusted now. Delivery is the part built for the current moment. Atlan ships an [MCP server](https://atlan.com/know/mcp-delivers-business-context/) today, model-agnostic by design, so a definition reaches an agent the same way regardless of which AI vendor's stack it runs on, the choice examined in [model-agnostic context layer](https://atlan.com/know/ai-agent/context-layer/model-agnostic-context-layer/) design. That contrasts with a tool built tightly around one vendor's stack, easier to fold into that roadmap than to keep serving a broader agent ecosystem alone. Teams sizing this decision should read [metadata tooling build vs buy evaluation criteria](https://atlan.com/know/ai-agent/context-layer/metadata-tooling-build-vs-buy-evaluation-criteria/) first. ### Core components of Atlan - **Enterprise Data Graph:** glossary, lineage, ownership, and policy connected into one governed graph - **Semantics and ontology:** business meaning generated and kept current, not a one-time pass - **Trusted, AI-ready data:** [data contracts](https://atlan.com/know/ai-agent/data-for-ai/data-contracts-for-ai/) and quality checks an agent can rely on before it acts - **Skills and procedures:** the operating knowledge for how work actually gets done, not just what data means - **MCP server:** model-agnostic delivery to any agent framework, not tied to one AI vendor's stack Get the CIO's Guide to Context Graphs A practical framework for evaluating a semantic tool, a platform upgrade, or a full context layer, before you commit to one. Get the CIO Guide --- ## Atlan vs Illumex: head-to-head Scoring the two head-to-head only makes sense if the comparison is honest about what each one was built to do, and what one of them no longer is. | Dimension | Illumex | Atlan | |---|---|---| | Primary focus | Generating business meaning from metadata automatically | Governing which meaning is authoritative and delivering it safely | | Availability | Not sold independently since the NVIDIA acquisition | Sold and deployed today, as an independent platform | | Key stakeholder | Data and analytics teams generating definitions | Governance, data platform, and AI teams jointly | | Delivery to agents | Tied to NVIDIA's own AI stack | MCP server, works the same regardless of model or vendor | | Governance after generation | Not part of its documented scope | Certification, ownership, and policy attached to every definition | | Measured by | Speed of generating a usable definition | Coverage of certified, owned, and current definitions across the estate | | Failure mode | A correct definition nobody keeps current once agents multiply | Governance metadata with no real generation engine underneath | Neither column is a knock on the other. [Full-stack AI platform vs. best-of-breed context layer](https://atlan.com/know/ai-agent/context-layer/full-stack-ai-platform-vs-best-of-breed-context-layer/) covers the general pattern NVIDIA followed here: a platform acquiring a well-executed point solution rather than reselling it standalone, one signal that generation alone was never going to be the whole answer, without that being a defect in what Illumex built. --- ## Is there still a decision to make now that Illumex is part of NVIDIA? Yes, just not the one the search query implies. There is no live bake-off between two competing vendors. The decision that remains is whether the category Illumex helped define, semantics generated automatically from metadata, is enough on its own from any vendor still selling it, or whether an enterprise running agents across many systems needs governance layered on top from day one. ### What NVIDIA's move signals for buyers evaluating this category A well-resourced acquirer paying real money for a narrow, well-executed semantic tool is evidence the underlying problem, meaning generated once and trusted, is worth solving. NVIDIA folding Illumex's team into its own R&D organization rather than continuing to sell the product independently is a second signal worth weighing.[2] That is not proof generation alone is insufficient, but it lines up with what [systems of semantics](https://atlan.com/know/ai-agent/data-for-ai/systems-of-semantics/) argues: semantics is becoming its own enterprise system, distinct from any single tool that generates it. Buyers comparing other semantics-first tools still on the market, [Cube](https://atlan.com/know/ai-agent/semantic-layer/cube-semantic-layer/), [Unity Catalog's semantic layer](https://atlan.com/know/ai-agent/semantic-layer/unity-catalog-semantic-layer/), or the players in [Lightdash vs dbt Semantic Layer](https://atlan.com/know/ai-agent/semantic-layer/lightdash-vs-dbt-semantic-layer-vs-context-layer/), are asking a version of the same question this page raises about Illumex. --- ## When would a point solution like Illumex have been enough, and when do you need a governed context layer? The right answer depends on how many systems and agents actually need the same definition, and who is accountable when one goes stale. **A narrow semantic-layer tool is enough when** one team generates and owns every definition in scope, a single AI stack consumes it, and informal review is sufficient change control. That was a real, common situation, and it is why Illumex found paying enterprise customers before the acquisition. **A governed context layer earns its place when** more than one team or agent depends on the same definitions, those agents run on more than one model, an auditor asks who approved a number, or "is this still current" needs an answer a person can stand behind. [Context layer for data governance teams](https://atlan.com/know/ai-agent/context-layer-for-data-governance-teams/) and [AI agent governance](https://atlan.com/know/ai-agent-governance/) describe that threshold in more depth than a comparison page needs to. **Plan for governance from day one when** the program is already multi-agent or spans several business units, since retrofitting ownership onto definitions three teams already depend on costs more than building it in early, the same argument [context versioning for AI agents](https://atlan.com/know/ai-agent/context-versioning-for-ai-agents/) makes about change control generally. Score your context maturity See how much of your business context is already generated, owned, and current enough for an agent to trust. Start the Assessment --- ## How Atlan approaches business context differently from Illumex Atlan's position is not that generating meaning from metadata was the wrong idea. It is that generation is the first half of a job that has a second half: keeping that meaning current, attaching an owner and a certification state to it, and delivering it through an interface that does not assume every agent runs on the same AI vendor's stack. The Enterprise Data Graph does the generation work Illumex did well, using [lineage](https://atlan.com/know/ai-agent/data-for-ai/automated-sql-lineage-vs-manual-lineage-mapping/) and usage patterns rather than requiring hand-written documentation, then attaches governance to the result instead of stopping at the definition. Delivery is where the two architectural choices matter most for a reader evaluating this today. Illumex's Generative Semantic Fabric was tied closely to NVIDIA's own AI stack well before the acquisition made that tie permanent. Atlan's MCP server was built the opposite way from the start, per [why MCP matters for AI agents](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/): a definition governed inside Atlan reaches an agent running on any model, not one vendor's roadmap. In Atlan's AI Labs benchmark, adding this governed context improved AI's text-to-SQL accuracy by 38%, a first-party result measured on context, similar in spirit to what [text-to-SQL for enterprise](https://atlan.com/know/ai-agent/data-for-ai/text-to-sql-for-enterprise/) finds breaks without governed definitions underneath. The same substrate connects to warehouse-native tools directly, as [Snowflake Semantic Views](https://atlan.com/know/snowflake/snowflake-semantic-views/) and [Unity Catalog Metrics](https://atlan.com/know/ai-agent/databricks/unity-catalog-metrics/) integrations show, a layering question that recurs in [Genie ontology and the Atlan context layer](https://atlan.com/know/ai-agent/databricks/genie-ontology-and-atlan-context-layer/). --- ## Real stories from real customers: One vocabulary, governed everywhere "Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs." — Joe DosSantos, VP Enterprise Data & Analytics, Workday Watch Now → "Atlan is much more than a catalog of catalogs. It's more of a context operating system…Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models." — Sridher Arumugham, Chief Data & Analytics Officer, DigiKey Watch Now → Neither company has published anything comparing Atlan to Illumex by name. Both describe, in their own words, the same second job this page argues Illumex never got the chance to finish: business language captured once, then governed and delivered to every agent that asks. See Atlan in Action: Live Context Layer Demos Watch how Atlan generates and governs business context as one part of a full context layer, not a standalone semantic tool. Watch the Live Demos --- ## The business-context question Illumex's acquisition doesn't answer NVIDIA paying a reported $60 million to $75 million for Illumex answers one question clearly: automatically generating business meaning from metadata is valuable enough that a major AI infrastructure company wanted it inside its own stack. It does not answer whether generation alone was ever sufficient for an enterprise running agents across dozens of systems, each with its own idea of what "active customer" means. NVIDIA folding the team in rather than reselling the product is a data point, not a verdict, but it lines up with the argument this page has made: meaning has to be generated, then governed and delivered the same way to every agent, regardless of which model that agent runs on. Book a Demo --- ## FAQs about Atlan vs Illumex 1. **What is Illumex, and is it still available as a product?** Illumex was an Israeli startup that built a Generative Semantic Fabric, software that inferred business definitions from metadata. NVIDIA acquired the company in February 2026, and it no longer sells the product as a standalone offering. 2. **What did Illumex's Generative Semantic Fabric actually do?** It read schema, column names, and usage metadata, never the underlying rows, and used that to infer what a table or field meant. Illumex packaged the result as a metric store, data dictionary, catalog, and business glossary sold together. 3. **Why did NVIDIA acquire Illumex, and what changed?** NVIDIA had already integrated Illumex's product with its NeMo platform in March 2025, and the acquisition, reported at $60 million to $75 million, formalized that relationship. Illumex's technology now lives inside NVIDIA's own agentic AI stack rather than as a standalone product. 4. **Is a semantic layer like Illumex the same as a context layer like Atlan?** No. A semantic layer generates and stores what data means. A context layer does that plus governs which definition is trustworthy, delivering it with trusted data and operating procedures through one interface. Illumex only ever sold the first job. 5. **Should I still evaluate Illumex, or look at Atlan instead?** Illumex is not purchasable as an independent product anymore, so there is no live bake-off. The useful question is whether a narrower semantic-layer point solution, from any vendor still selling one, covers enough of the job, or whether the estate needs governance across every system. 6. **Can Atlan do what Illumex did?** Atlan's Enterprise Data Graph generates and maintains business definitions from metadata and lineage, the same core job Illumex marketed as Business Context AI, then adds certification and policy on top, delivered through a model-agnostic MCP server. --- ## Sources 1. [Illumex raises $13M to add more meaning and context to structured data used by generative AI, SiliconAngle](https://siliconangle.com/2024/06/27/illumex-raises-13m-add-meaning-context-structured-data-used-generative-ai/) 2. [Nvidia acquires Israeli data co Illumex, The Jerusalem Post](https://www.jpost.com/business-and-innovation/tech-and-start-ups/article-887707) 3. [Gartner Says Lack of Semantics Causes Inaccurate AI Agents and Wasted Spending, Gartner](https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-says-lack-of-semantics-causes-inaccurate-artificial-intelligence-agents-and-wasted-spending) 4. [Nvidia acquires Israeli data co Illumex, Globes](https://en.globes.co.il/en/article-nvidia-acquires-israeli-data-co-illumex-1001535871) 5. [illumex Generative Semantic Fabric, AWS Marketplace](https://aws.amazon.com/marketplace/pp/prodview-7iity5oddyx36) 6. [How Illumex and NVIDIA are partnering to address the trust problem with GenAI and agents, R&D World](https://www.rdworldonline.com/how-illumex-and-nvidia-are-partnering-to-address-the-trust-problem-with-genai-and-agents/)