The questionIs this a cloud service we connect to?
AiDN is not a service the operator reaches out to. It is a containerised platform that deploys where the operator's data already lives — on-prem, private cloud, or hybrid where only model inference touches an external API. The reference deployment is the same architecture running in production today, not a simplified demo variant.
What we can evidence
Deploy the production architecture, not a demo variant.
The reference deployment is a containerised stack on Docker Compose running on a single Linux server inside the operator's network — the same architecture running in production across operator networks on five continents today.
Connect to OSS and BSS without a replication pipeline.
Data-plane connections are read-only and stay entirely internal, with jump-server routing where network segmentation requires it. Agents query data in place, so no data lake replication is needed and no firewall exception is opened for an external service to reach in.
Work in environments with no route to the public internet.
Proven in constrained operator environments where client machines cannot reach public content delivery networks — every front-end asset is served locally from the platform host, so the user interface has no external dependency at all.
The direction of the connection is the whole argument
The market is beginning to distinguish between two architectures that sound similar and behave very differently: AI that lives in the cloud and reaches into the operator, and AI that lives in the operator's estate and reaches out only for compute when needed. The first requires firewall exceptions, replication pipelines, or both — architectural debt that accumulates quickly and is difficult to unwind. AiDN is built on the second model.
What actually gets deployed
A containerised stack on your infrastructure
Docker Compose on a single Linux server inside your network is the reference footprint. It scales out from there, but it does not need to start large.
Read-only connections to your systems
Scoped, read-only credentials against the views and tables you nominate. Purpose-built query tools per source — agents do not get free-form SQL against production.
Local retrieval and indexing
RAG, embedding generation and vector search all run on the platform host, so even semantic indexing of internal documents needs no external service.
Outbound HTTPS for inference only, if you choose it
In the hybrid default, the only traffic leaving your perimeter is a model API call carrying a prompt and query results. In the fully sovereign configuration, nothing leaves at all.
Hybrid or fully sovereign — decided once, at deployment
| Configuration | What leaves the perimeter | When to choose it |
|---|---|---|
| Hybrid (default) | Prompts and query results to a frontier model API, under terms excluding training | Most operators — best reasoning quality with contractual and architectural protection |
| Fully sovereign | Nothing | Where policy or jurisdiction requires absolute containment |
The choice is a configuration decision made with your security and compliance teams, not a different product.