Pinned Tweet
Introducing CData Connect AI Gateway.
One gateway between your AI and everything it touches. @ravivlevi 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 👇
You’ve spent years teaching people how your company works. Your AI needs that knowledge too.
Introducing Connect AI Gateway.
It routes, governs, and learns from AI requests, building a system of record for how your company thinks.
Early access is live: cdata.com/ai
Read-only in two places. In our testing, a @ChatGPT dot with its plugin set to “Allow read-only tools” still created an @Asana task when we asked for one directly.
If a dot must never change a system, set read-only on the connection, too: bit.ly/4z2BwJW
1
54
Five years on the Deloitte Technology Fast 500.
Thank you to the customers, partners, and team who took CData from connecting enterprise data to governing how AI works in the systems that run the business.
Details: bit.ly/4ACBNof
Three questions for any MCP gateway:
What does a correct task cost on your data?
Where do joins and calculations happen, in the gateway or the model?
Is business context served on demand or in every prompt?
Our benchmark: bit.ly/4sAivx4
37
Agent 365 controls what agents can do. CData Connect AI manages the secure data access behind them.
A single managed MCP connection links Copilot Studio agents to hundreds of enterprise systems.
CData is at Microsoft Ignite, booth 5241R.
See how Copilot Studio agents connect: bit.ly/4kkp2cV
Nov. 17-20 | Moscone, San Francisco
#MSIgnite
5
110
Great customer support starts with great data access.
@Edsal_Mfg gave Ava, their AI agent, live access to SQL Server data. Order status. Shipment information. Parts lists. The same information a human support agent has.
Read the full Edsal case study: bit.ly/4sd3NMi
1
1
84
Foundations 2026 is now on demand. Seven sessions on the AI gateway: the case for it, how it works, how it learns your business, and what a CIO’s rollout really looks like.
Free to watch, no registration: bit.ly/4k9w3gw
$0.007. That’s what an agent spent answering “Which accounts renew this month?” from live company data. The answer came back correct.
@PrimaLabsAI ran an open-weight model on low-cost inference. CData Connect AI Gateway connected the agent to the data, applied the user’s permissions, and logged the request.
One test, so costs will vary.
Full write-up and how to try it: bit.ly/4iEc4Wp
1
2
75
CData retweeted
AI gets expensive and risky in the same two ways.
It burns a heavy model when a light one would do.
And it grabs data it was never meant to touch.
Most people never catch it. They pay the bill and hope the access rules held up.
I've run businesses across a stack of tools for years.
My data sits in Salesforce, in NetSuite, in a dozen other places.
And every time I asked an AI something real, it either guessed at what my terms meant or pulled way more than it needed.
CData Software showed me their Connect AI Gateway. It deals with both problems before the model ever runs.
Watch one prompt go through it. "What revenue is at risk for mid-market in Q3."
The gateway reads the intent first. It's just a lookup, so it sends it to a lighter model instead of the pricey one.
Then it works out what my words actually mean. Mid-market. This quarter. Revenue. At risk. Each one maps back to a real system or a real definition.
Then it pulls only the records I'm allowed to see. The permissions and the masking happen in the gateway, not the model.
And it hands the model one clean subset. Not a data dump.
That last part got me. The raw pull was 200,000 tokens. What reached the model was 9,500. Same answer, a fraction of the cost.
Every step logged too. Prompt, permission, record, response.
Here's the part I keep thinking about. When you confirm what "at risk" means, that sticks. It becomes a saved fact. The thing knows a little more than it did a minute ago.
Everyone chases the flashy AI stuff. But the boring layer that controls cost and access is what makes the rest safe to actually use.
Check out CData here: cdata.com/ai/
P.S. Building a 10-business portfolio. 6 businesses in, all remote.
#CDatapartner
A wrong write lands in a live system. Unauthorized writes showed up in 29% of runs through an open-source MCP gateway (17 models) vs. 10% through Connect AI Gateway (5 models), with the same business context.
Details: bit.ly/4sAivx4
2
63
CData retweeted
The fastest open models, now connected to live enterprise data.
We're excited to partner with @cdatasoftware . PrimaLabs brings fast open-model inference, ranked at the top of Artificial Analysis.
CData Connect AI Gateway brings governed access to hundreds of enterprise sources.
Together, teams can run agents on open models with the data access and controls they already need.
Looking forward to building this with the @cdatasoftware team
cdata.com/blog/cdata-primala…
Four ways to connect @ChatGPT, ChatGPT dots, or @Meta Muse (@AIatMeta) to business systems through MCP.
They differ most once you have several connections: who manages them, where the log lives, and who sets permissions.
Our buyer’s guide compares them by use case: bit.ly/4z2BwJW
James Rochette joins CData as Chief Financial Officer, and June Yang, formerly of Google Cloud, joins our Board of Directors.
Two additions to help us scale.
Full announcement: bit.ly/3UKBwAk
4
85
16 of 27 models completed every task through Connect AI Gateway. Through an open-source MCP gateway with the same business context, 3 did, all frontier. Economy models: 78% of tasks vs. 17%.
Full benchmark: bit.ly/4sAivx4
4
114
What revenue is at risk for mid-market in Q3?
One ordinary question. To answer it, an agent needs a model, accounts from Salesforce, revenue from NetSuite, the company’s definition of “at risk,” the right permissions, and a way to act on the result.
Life of a Prompt follows the question through the Connect AI Gateway, where all of it comes together in a single run: bit.ly/48OdQ17
In the CData Arc Q4 2026 release, the built-in Model Context Protocol (MCP) server can take action. Ask it to run a connector, requeue a failed message, or clean up old files. Every action requires explicit confirmation and is recorded in the audit log.
Plus one-click AI diagnosis for failed transactions.
Details: bit.ly/4jQ0ybm
.@Adobe’s AI agents each get exactly the data their use case requires. Toolkits handle the scoping.
Cloud Gateway bridges on-premises systems without firewall changes. All of it runs through CData Connect AI: bit.ly/4kce74Z
CData retweeted
I built the world's first MCP server in Doom, and I've had so much fun showing it off at the CData booth at AgentCon + MCPCon in San Jose this week!
Customers don’t follow scripts. Great support moves with the conversation.
Art Courtney of @Edsal_Mfg explains why flowing with the context of a call makes Ava, their AI agent, work. He also shares how to build it.
Ava is built on CData Connect AI. Hear Art’s approach and get the details: bit.ly/4sd3NMi
1
69














