Give an AI agent access to everything and it gets worse. That is the uncomfortable engineering reality behind many stalled enterprise AI deployments, and it is the specific technical problem Modus emerged from stealth this week to solve. Axios reported that the company announced a $10 million seed round led by Insight Partners, with participation from Soma Capital, Bullet Ventures, and technology founders and operators including Eyal Kishon, Nadav Avrami of Wix and Dazl, the co-founders of Cyera, and the founders of Epsagon.
Modus calls its product the Context Warehouse. Understanding why it exists requires understanding how enterprise AI actually fails.
The Failure Mode
The intuitive approach to making AI useful inside a company is to connect it to more systems. In practice, enterprises are discovering that more connections often produce slower, more expensive, and less reliable results. Agents over-fetch information. They re-query enterprise systems repeatedly. They consume far more tokens than a task requires, because nothing in their architecture tells them which context actually matters for the question at hand.
The consequences compound. Costs climb. Answers cannot be trusted. Pilots stall before they reach production. Modus describes the gap between what AI can access and how the business actually works as the Context Gap, and its diagnosis inverts the usual assumption. The problem is not too little context. It is too much of the wrong context.
There is a deeper technical point underneath. A model can access tables in a warehouse, dashboards in BI tools, documents, tickets, code repositories, and collaboration systems. What it cannot inherently understand is which definitions the business trusts, which dashboard teams actually rely on, why a metric changed last quarter, or which business logic should take precedence today. Access is not understanding.
How the Context Warehouse Works
The Context Warehouse is a continuously learning layer that sits independently of any data warehouse, AI model, or application platform. Instead of relying on documentation, semantic models, or application-specific context that must be manually maintained as the business evolves, it learns how an organization actually operates from real usage.
Its inputs span structured data, unstructured data, and the tribal knowledge in between: the queries analysts keep coming back to, the dashboards teams rely on, the pipelines, docs, and decision threads that trace how the business really runs. The design principle is that context is earned from real usage, not declared once and left to drift.
At inference time, the system composes only the relevant context required for each AI interaction. Agents reason on signal instead of noise. The company reports that this reduces unnecessary retrieval and token consumption by up to 10x.
The architecture avoids centralization deliberately. The platform learns from metadata and usage patterns across data warehouses, BI tools, pipelines, code repositories, documentation, and collaboration systems, without requiring organizations to pull sensitive business data into one place. The reasoning: context cannot live inside one platform if the business does not live inside one platform. Integration with existing agent stacks happens through the tools teams already use, including MCP.
Governance at the Context Layer
Security is handled before the model ever sees anything. Governance is enforced before context reaches the model, so every AI interaction receives only the information it is authorized to access. Sensitive customer data remains inside the customer’s environment. The framing from Modus is blunt: agents are only as safe as the context they receive.
Why the Founders Built It
Modus was founded by Daniel Shimoni and Tomer Mesika, and both encountered the problem before naming it. As VP of Product at Lusha, Shimoni saw that AI was only as useful as the business context behind the data. As Head of Architecture at Cyera, Mesika built infrastructure to classify, govern, and secure enterprise information at scale. Their conclusion was the same: the systems enterprises rely on were never built for AI agents.
“Companies are no longer just trying to get their teams to use AI. They are asking how to scale it across the organization without accuracy dropping, governance breaking, or costs spiraling,” said Shimoni, the company’s CEO. “Whether people call it a company brain, a context layer, or context engineering, they are all trying to solve the same problem. We believe every enterprise needs a continuously maintained understanding of how the business operates before it can build any of those things. That is what the Context Warehouse provides.”
The Maintenance Problem Nobody Budgets For
Plenty of engineering teams have already built internal context layers. The harder question is what those systems look like a year later, after reorgs, metric redefinitions, new tools, and changed business logic.
“Building a context layer is not the hardest part,” said Mesika, the CTO. “Keeping it current is. Every change your business makes changes the context AI depends on. The real decision is no longer buy versus build. It is whether you want to own the ongoing cost of maintaining that understanding. We built the Context Warehouse so engineering teams can build what differentiates their business instead of maintaining the infrastructure underneath it.”
The contrast with traditional infrastructure is part of the pitch. A data warehouse takes years of pipelines to fill and maintain. The Context Warehouse fills itself, learning from the work already happening.
Early Results
The platform is already running with enterprise customers across financial services, technology, and SaaS. Reported outcomes include improved AI accuracy, stronger governance, faster response times, and lower costs of operating AI at scale.
Insight Partners sees a category forming. “Every major wave of enterprise software has required a new foundation,” said Ganesh Bell, Managing Director at the firm. “Data warehouses became foundational infrastructure for enterprise data. As AI becomes production infrastructure, organizations need a system of understanding that every agent and application can build on. We believe Modus is defining that category with the Context Warehouse.”
The roadmap points beyond retrieval. Over time, Modus intends the same foundation to support a more proactive class of AI, one that surfaces what matters, detects what changed, and helps teams move from trusted answers to trusted action.
