Two decades ago, we started CData with a belief that data should be accessible to anyone who needs it, in whatever tool they already use. Every step since has moved toward that goal: connectors, then a data access and integration platform, and now governed access to enterprise data for AI agents.
We believe our move to Challenger in the 2026 Gartner® Magic Quadrant™ for Data Integration Tools is a sign we're on the right track, because the work behind it is the same work that makes an AI gateway possible.
It started with a belief about access
Making data accessible meant solving the problem nobody wanted: every source speaks its own dialect. So we started with connectors, standards-based access to any source from any tool. Many teams spend months building connectors before a project even starts, and we wanted that work to disappear. Hundreds of sources later, the lesson wasn't breadth. It was depth. Every system has its own custom fields, query limits, and quirks, and an application only works well when the connector understands them.
Two decades of that connector engineering is now a body of knowledge about how enterprise systems actually behave. It turns out to be what AI agents need most.
Then the job grew
Accessible data then led us to data integration, with every capability it takes: replication with change data capture (CDC) to keep warehouses and applications current, federation and virtualization to query sources in place without copying anything, and partner exchange for data that crosses company lines. These capabilities share one platform, so the same connections serve a dashboard, a pipeline, and a trading partner.
When the Model Context Protocol (MCP) arrived, that same foundation made the next step natural. We were one of the first to build MCP servers, including managed MCP servers, so AI tools could reach enterprise data with governed access, through the connections and depth we had already built.
Then the consumer changed
Our customers still need the same three things: connectivity to the right data, context for what it means, and control over who uses it. What changed is who asks for it. Agents now sit alongside people and applications as consumers of enterprise data, and they ask continuously and across systems, with no one reviewing a wrong column before it becomes an answer.
Two other shifts arrived at the same time. Connector counts stopped separating vendors, and the models started to look the same. What's left is context: what the data means, who defined it, and what an agent is allowed to do with it. Context can't live in the model, because models get swapped, or in each agent, because each would learn it separately. It has to live in the layer every request passes through, between agents and the systems they work with.
Then we bet on the AI gateway
Our customers and the market were starting to ask for exactly that: one layer between agents and their data, governed from a single control plane. We checked it against the foundation we'd spent two decades building, realized the AI gateway was the next move, and made a bet on it. Launched this fall, CData Connect AI Gateway extends that governed data layer to agents in three ways:
Connectivity: Agents use the same connections and depth, so they reach custom fields and not just standard objects.
Context: The Context Engine learns what the business means from system knowledge, documented definitions, and the unwritten rules people carry, so agents give accurate answers while spending fewer tokens. Ask about "this quarter," and it resolves correctly because it knows the fiscal year starts in February.
Control: One control plane enforces policy before data reaches a person or an agent, ties each request to the real caller, and logs every step, so traffic can grow without the risk growing with it.
The gateway sits in the traffic path between agents and systems of record, which is the only place context can be learned continuously.
Where this goes
The data layer holds what a business knows, and the AI gateway governs every request that uses it. Each request teaches the layer something, so the two sharpen each other. That loop, and not the model, is where the enterprise AI race gets decided.
Read the full 2026 Gartner® Magic Quadrant™ for Data Integration Tools report.
