Try Connect AI Gateway at no cost today
We are accepting a limited number of early access program participants to try Connect AI’s LLM gateway and context engine capabilities. Tell us about your needs and environment, and we’ll get back to you within two business days if you’re a fit.
No credit card. By submitting, you agree to CData’s Privacy Policy.
97.6%
fewer tokens spent when the gateway returns the answer set instead of the full record set.
14 days
free trial access to all Connect AI Gateway capabilities including MCP connectivity to hundreds of systems.
No cost
for LLM gateway & context engine capabilities through the end of 2026, even after the full trial concludes.
What you can do with Connect AI Gateway
Most agents pull every matching record into the prompt and let the model sort it out. Connect AI runs the query at the source and returns only the answer set.
"Which open opportunities over $250K have had no activity in 30 days?"
source salesforce.Opportunity
rows read 18,402
rows sent 37
context −97.6%
billed to team: revenue-ops
agent: pipeline-review
Why it matters: Spend drops because the model reads less, and every token is attributed to a team, agent, and model.
Most organizations use a shared service account that gives every agent the same reach. Connect AI carries a delegated user identity on each request and enforces it inside the source at runtime.
POST /v1/query On-Behalf-Of: [email protected] Agent: support-triage SELECT * FROM Zendesk.Tickets WHERE Priority = 'urgent'
identity [email protected] policy zendesk:region = EMEA records 12 of 41 visible decision allowed (row filter) trace req_8f21c4
Why it matters: Agents see only what the person they act for can see, with a trail behind every answer.
When business logic lives in the prompt, you are locked to the model that can follow it. Moved to the data layer, model choice becomes a cost decision.
routes:
- match: task == "summarize"
model: claude-haiku
- match: task == "analyze"
model: gpt-5
fallback: claude-sonnet
req_1a90 summarize → claude-haiku
req_1a91 analyze → gpt-5
req_1a92 analyze → gpt-5 (timeout)
fallback → claude-sonnet
Why it matters: Each prompt goes to the model that fits it, and failover happens without an application change.
Teams typically stand up MCP servers one at a time, each with its own auth and no shared audit. Connect AI includes hundreds of proven first-party connectors and can manage any MCP server.
{
"mcpServers": {
"cdata": {
"url": "https://gateway.cdata.com/mcp",
"auth": "oauth"
}
}
}
netsuite.query managed snowflake.query managed servicenow.incidents managed internal-hr (yours) registered auth sso · audit on
Why it matters: One place for authentication, permissions, and audit across every MCP server your agents call.