CData Software

Since our beginning, CData has done one critical thing better than anyone: connect to enterprise systems and understand what’s actually inside them.     Every object, every relationship, every validation rule, every custom field no vendor knows exists. That work became the data layer embedded in Palantir Technologies, Google, and hundreds of enterprise products, and the connectivity more than 10,000 companies run in production today.     For years this has been important. With enterprise AI, this data layer is now non-negotiable.     In this article, we take you inside Connect AI Gateway. You'll learn how deep enterprise MCP connectivity, company-wide context, governance controls, and smart model routing enable any IT leader to answer the three most pressing questions facing AI rollouts:     • What is all of this costing?     • What can our agents see and do?     • Can we trust the answers and actions?

A context layer inside the AI gateway learns from every question. It sees the data that comes back. It sees whether the answer held up.     That idea is at the heart of Raviv Levi's session at The AI Conference today, "What most AI gateways and goldfish have in common." Our CPTO will cover where your context layer can live and what opens up once it's inside: lower token spend, sharper agent accuracy, and read-write use cases teams had set aside.     It's the thinking behind the Connect AI Gateway we launched this week. Bring your questions. The team is at booth 132.     Sept. 30 | 1:30p.m. PT | Theater 4 

Ten years ago, Raviv Levi had a front-row seat at Cisco as enterprises moved their network traffic through a new kind of gateway. The questions came up in the same order everywhere: cost first, then security, then control.    AI is following the same path. In his new blog, our CPTO explains why we built CData Connect AI Gateway, one control point for enterprise AI, and where it goes next. 

Introducing CData Connect AI Gateway. One gateway between your AI and everything it touches. Raviv Levi and Amit Sharma show what we're announcing today.   Every question your teams and agents ask now runs through a layer that knows your systems, your permissions, and how your business works. The answers are accurate, the actions are reliable, and the token cost drops to a fraction of what it was.   Any model. Any platform. One context layer that goes with you.   Watch Raviv and Amit walk through it 👇

A year ago, the question was how to connect AI to your systems. Now agents are working inside them, and the questions have changed: Is the answer right? Whose permissions applied? What did each request cost?     CData Connect AI Gateway, launching today, answers all of them from one control point. It grounds every request from your teams and agents in company context, governs it down to the record, and routes it to the most efficient model for the task.     It's built on the data layer behind Palantir Technologies, Google, and hundreds of enterprise products, used by thousands of organizations. That's why it understands the systems on the other end of every request from day one.     Early access opens today. Amit Sharma explains why the gateway has to start at the data. Link in the comments. 

Every result in the 178x Spread benchmark is reproducible. The test harness, the prompt set, and all 1,034 runs are published in the public repository. If you want to know how your specific models and systems perform, you can run the same methodology against your own data. The benchmark compared 22 models, economy to frontier, against live enterprise data through Connect AI. The findings on accuracy, safety, and cost are in the full paper, linked in the comments.

Calling an AI model is easy.    Letting an entire company use AI without blowing the budget, losing control, or giving the compliance team heartburn? That’s where it gets interesting.    👋 Joe Karlsson gives a quick overview of AI gateways, and why they’re becoming essential for scaling AI across the business.

Foundations returns November 5 with one focus: the AI gateway. See how an AI gateway can ground every agent and AI tool in your business from day one, enforce governance down to the record, and get smarter with every interaction. Hear from CData CEO Amit Sharma and CPTO Raviv Levi on why AI gateways and enterprise MCP platforms belong together. Go inside the self-learning context engine that powers it, then hear from Erik Bailey, CIO at Anaqua, on putting AI to work today. It’s the biggest Foundations yet.

On raw access to enterprise data, not one of 22 AI models stayed within what it was authorized to change across every run. The worst run queued 628 rows it was never permitted to touch. The frontier models were among the worst offenders. When write-time validation moved into the governed data layer, so the tool itself checked every change against the rule for what was eligible, unauthorized writes dropped to zero. On that guarded build, 21 of 22 models were correct with zero unauthorized writes. Safety is a property of the architecture, not the model. The data layer is where it has to live. Full benchmark from CData Labs in the comments. 

Where enterprise agents are spending their time. We looked at where agents actually spend their time once customers connect them to business data, and one pattern over the past few months stood out: a lot of the activity moved toward ERP and CRM. Sage Intacct, Salesforce, NetSuite, and HubSpot sat at the top of the systems in this view. That makes sense when you think about the questions people are asking. What's still outstanding? Is the period ready to close? Where does this deal stand? These are operational questions that depend on what's true right now, so the answer usually lives in the system the business is already running on. This looks like a new access pattern alongside analytics and the warehouse, centered on live operational data. The systems businesses run on are increasingly becoming the systems their agents work in too.