For your AI agents
Through MCP or HTTP
Agents query databases through controlled access. Identity, permissions, result policies and any required approval are evaluated in the same chain.
Sentinel by NewHeap Pilot preview
All applications running on a supported SQL Server or PostgreSQL database can become AI-ready for your organization with Sentinel. Your AI agents use their data through controlled access.
No new or additional application development.
We configure the gateway and access policies. For supported database versions, drivers and features. Pilot preview, not yet released for production.
“What is the revenue in the North region?”
Your existing application keeps working
| Order | Region | Revenue |
|---|---|---|
| 1048 | North | €12,400 |
| 1049 | North | €8,200 |
| 1050 | North | €4,200 |
Your existing software and database. No additional application development.
Sentinel checks the request and which data may be returned.
What is the revenue in the North region?
€24,800
Revenue · North region
You may read revenue for your own region.
Which customers placed orders this month?
Customer names are pseudonymized; email addresses remain protected.
What do the employees earn?
Your Sales role does not grant access to salary data.
Only released information reaches the user and the selected AI model.
| Request | Identity | Decision | Applied policy |
|---|---|---|---|
| REQ-01 | user-07 · Sales | Allowed | Read permission |
| REQ-02 | user-07 · Sales | Protected | Field protection |
| REQ-03 | user-07 · Sales | Denied | Role boundary |
The audit log stores decisions and metadata, not question text, SQL, parameter values or result data.
Your existing software takes on a new role
Your organization already uses applications for orders, customers or inventory. Sentinel makes data in their supported databases available to AI agents without new or additional application development. An agent can then prepare a report after a meeting using current business data. You decide which information the agent may use and which actions require approval.
Define which tables and columns each application and user may access. Sentinel evaluates the complete query, including supported queries that combine several tables.
The gateway checks results against configured policy. Configured source columns are also recognized through aliases and supported database views.
Reading data and changing it have different consequences. Where policy requires it, a supported write waits for separate human approval.
The audit log shows who made a request, which rules applied and what was decided. The protected content itself is excluded from this evidence log.
Two ways to connect
For your AI agents
Agents query databases through controlled access. Identity, permissions, result policies and any required approval are evaluated in the same chain.
For existing applications
The application keeps its existing database and driver. For supported features, only the connection address changes to point to the gateway. Sentinel makes permitted data available to AI; this requires no application development.
Deployment with your IT team
In the customer-hosted preview, the connector and database gateway run within your network. Database credentials and raw results stay there; only results released according to policy reach the requester.
The NewHeap-hosted variant is still in development and is not available as a production service.
A focused pilot
Sentinel is a pilot preview by NewHeap. Supported features have been tested in development and pilot environments; there is no production release yet. Together, we define the database, access policies, test scenarios and acceptance criteria for your use case.
You retain ownership, access and control.
Processing varies by service and selected model.
Security is part of the foundation.
Choose suitable models, with Local AI where needed.