On-prem data, cloud AI: a no-trade-off architecture
For many enterprises, the data that matters most lives behind a firewall. Financials, customer PII, manufacturing telemetry. Moving any of it to the cloud is a non-starter — too much risk, too much paperwork, sometimes outright forbidden by regulation.
The traditional response was to forgo modern AI altogether. We don't think that's the right trade.
The Gateway model
AI Insights ships with a small Windows agent that runs inside your network. The agent does three things:
- Polls the cloud for queries (outbound HTTPS only — no inbound firewall rule).
- Runs each query locally against your on-prem SQL Server, PostgreSQL, or MySQL using credentials that never leave your network.
- Posts the result rows back to the cloud over the same outbound channel.
The cloud-hosted AI agent does what it does best — natural-language understanding, SQL generation, chart construction. Your data does what it must — stays put.
Why this works
- Outbound-only: zero changes to your firewall ingress rules.
- Credentialed locally: the connection string never crosses the WAN unencrypted.
- Auditable: every dispatched query is logged on the gateway machine.
- Revocable: rotate the gateway token in one click; the cloud can no longer dispatch.
Where you'd use it
- A bank running on-prem SQL Server for customer accounts
- A hospital with patient records in a HIPAA-zoned SQL warehouse
- A manufacturer with operational technology data behind an air gap
In every case, AI Insights gives the same conversational experience the cloud-native users get — without the data ever leaving its rightful home.