CData Software Inc. is expanding its Connect AI platform with new tooling and governance features designed to help enterprises move artificial intelligence systems from experimental pilots into production environments.

The announcement at Gartner Data & Analytics Summit in Orlando adds new capabilities across connectivity, contextual intelligence and security controls. The company says those three elements are critical for enabling AI agents to safely interact with live enterprise data and execute real-world business tasks.

Though enterprise spending on AI is accelerating, many initiatives still fail to reach production because the underlying data infrastructure cannot support them, CData executives said.

ConnectAI provides real-time, governed access to more than 350 enterprise data sources using the Model Context Protocol, a standard framework that lets AI systems securely connect to and interact with external data sources and applications such as Salesforce Inc. customer relationship management, SAP SE enterprise resource planning and various databases. Those connectors are now being repurposed for AI systems that need direct access to operational data.

CData said a key barrier to moving AI into production at many companies is that organizations are still wiring together custom integrations or moving data into separate systems before AI can access it.

“Companies’ ability to have success with AI is not limited by the models that they’re leveraging,” said Will Davis, CData’s chief marketing officer. “It’s limited how they’ve established data infrastructure for AI specifically.”

The new release expands Connect AI in three areas: connectivity, context and control.

For connectivity, a new gateway capability extends live read/write access to enterprise data sources behind corporate firewalls. CData uses a standardized relational interface that sits on top of the source systems and allows querying of multiple data sources using standard SQL. That enables AI agents to query and interact with multiple enterprise systems through a common interface, allowing them to explore relationships across data sources and perform complex operations.

“It’s an SQL abstraction layer,” said Marie Forshaw, CData’s senior vice president of product marketing.

Agent tooling

The announcement also introduces three categories of “agent tools” that determine what actions AI systems can perform.

Universal tools provide a standardized set of operations that work across all connected systems. Source tools expose system-specific functions, such as creating a lead in Salesforce or opening a ticket in Jira. Custom tools allow organizations to build purpose-built operations tailored to particular workflows.

“Universal tools look like database functions for such actions as insert and update,” Forshaw said. “They provide a lot of versatility across any source system.”

CData says one of its strengths is its ability to support complex queries that befuddle competing platforms. In its own tests of 378 prompts across enterprise systems such as customer relationship management and enterprise resource planning, CData said, its platform achieved 98.5% accuracy compared with 65% to 75% for competing MCP implementations. Errors were most common in areas such as relative date logic, multiple filters and semantic interpretation of business terms.

Query handling becomes significant when AI agents execute multistep workflows, said Jerod Johnson, CData’s director of technology evangelism. “The engineering and architecting work that we’ve done to add connectivity, context and control to our MCP platform makes models more accurate than other providers’,” Johnson said.

For governance, Connect AI now supports cross-domain identity lifecycle management to allow organizations to enforce authentication and authorization using existing source-system permissions.

Because the platform connects directly to live systems, Forshaw said, user permissions can carry over automatically when AI tools access enterprise data. “I know permissions and restrictions are going to be respected and flow through to when people use [Anthropic PBC’s] Claude to access Salesforce data,” she said.

Executives said CData is focused on supporting a shift from experimentation to production deployments built on governed data infrastructure.

CData has raised over $510 million in venture funding and has a nine-figure annual recurring revenue rate, Davis said.

Image: Shutterstock