CData’s AI gateway governs agents’ access to enterprise data

Data integration firm CData Software Inc. unveiled Connect AI Gateway, a tool aimed at regulating how AI agents access data and make decisions. The product, currently in early access, builds upon CData's existing Connect AI platform and its Model Context Protocol. The gateway consolidates the registration of models, MCP servers, and agents, allowing IT teams to establish rules for their activities.

A pivotal feature is the context engine, which utilizes connected systems' structure, business definitions, and user interactions. This results in more accurate AI answers, curtails data exposure, and manages model expenses. CData already offers connectors to numerous business applications and databases. The gateway extends these connections to enforce user permissions when agents retrieve data or perform actions.

Policies can control access to specific data, with an audit trail documenting prompt, model, tool, and policy choices leading to a response. The company emphasizes that an agent's answer can vary depending on the business definitions used, such as the concept of "revenue." The gateway captures less formal knowledge from documents, conversations, and response corrections, storing this in a context graph outside individual AI models.

This allows organizations to apply shared context when changing models. IT teams can review the graph to determine what should be shared. The gateway manages separate routing and control functions for models, MCP tools, and agents, enabling token budget allocation and request routing based on policies. This allows companies to select the most suitable model for a task based on factors like features and cost.

CData plans to automate the selection of the cheapest model based on policy, though the ultimate decision will be intelligent and consider system behavior. Additionally, the data layer can reduce workloads on models, saving on token costs by filtering, joining, and aggregating records before sending data to the model. In a test involving 378 enterprise queries, Connect AI answered 98.5% correctly, compared to 65% to 75% for other MCP providers tested by CData.

The gateway represents a shift in CData's focus, moving from giving AI tools governed access to enterprise systems to overseeing the entire process from prompt to action. The context engine complements existing data management tools but is not intended to replace a full data catalog or semantic layer.