CData Software has introduced Connect AI Gateway, a platform that governs how artificial intelligence agents select models, use tools and access enterprise data. The product entered early access on September 29, 2026, as an extension of the company’s Connect AI platform and managed Model Context Protocol offering.
One control layer for agents
The gateway lets IT teams register models, MCP servers and agents in one place, then apply rules to their capabilities. Policies can enforce a person’s existing permissions down to specific database rows and columns. An audit trail records the prompt, model, tool and policy involved in an agent response or action.
CData says the approach is intended to cover the full request path, from the initial prompt through model selection, data retrieval and any resulting action. That gives administrators a way to investigate incorrect answers or blocked operations without treating model access as the only control point.
Context beyond the model
A context engine uses information about connected systems, business definitions and user interactions to shape agent responses. CData says it can reuse definitions maintained in Fivetran’s dbt and Microsoft Power BI, together with field and relationship information from source systems.
The system can also retain knowledge from documents, conversations and corrections in a context graph outside individual AI models. People review that graph and decide which lessons become shared organizational context. CData says the gateway is not meant to replace a full data catalog or semantic layer, and the initial release cannot directly import definitions from data catalogs.
Routing and cost controls
Separate controls govern model, MCP tool and agent routing. Organizations can set token budgets and choose models according to factors such as features and cost. Automatic selection of the cheapest suitable model is planned, but the initial routing approach is policy-based.
CData also says its data layer can filter, join and aggregate records before sending results to a model, potentially reducing the amount of information placed in its context window. The company’s stated goal is to limit both data exposure and token spending.
Company-reported results
In a company-run test of 378 enterprise queries, CData said Connect AI answered 98.5% correctly, compared with 65% to 75% for other MCP providers it tested. CData also reported a cost difference of up to 175-fold between models that produced the same correct answer. These figures are company benchmarks, not independent tests or documented savings from a customer deployment.
The context graph introduces an operational choice for customers: a correction that helps one employee may mislead others. Human approval of what becomes shared knowledge is therefore part of the product’s governance model.
AI-generated editorial illustration; not a documentary photograph.
Source: CData’s AI gateway governs agents’ access to enterprise data
SIGNAL TITANS
Leave a Reply