Why we built Nexus AI on top of Smallworld GNM instead of around it

The conventional approach to adding AI to a Smallworld Geo Network Management (GNM) environment is to export data, run it through a pipeline, and build something that reads from that pipeline. It’s the approach most vendors take. It’s also the approach that creates the most problems.

The export problem

When you export Smallworld GNM data to feed another system, you get a snapshot. The moment the export finishes, the data starts aging. For operational use cases — trace requests, quality checks, bulk updates — you need data that reflects what’s actually in the network model, not what was in it yesterday.

Building on top of an export also means building and maintaining the export. When Smallworld GNM is upgraded, the export breaks or drifts. When data models change, the pipeline needs updating. The maintenance burden compounds over time.

What “native to Smallworld GNM” actually means

Nexus AI products are written in Magik, against the Smallworld GNM API. That’s not a marketing statement — it’s a consequence of the architecture.

It means the Assistant reads your network model directly. It means the Controller manages Smallworld GNM processes from inside the process boundary. It means the calculations in Workspace run against live data, not exported data.

The result is a platform that behaves like part of Smallworld GNM, because it is.

The privacy decision

Utility network data is critical infrastructure. We made an explicit decision: Nexus AI products are designed so that sensitive network data never leaves the operator’s environment to train an external model. The AI sees command logic, not your network topology.

That decision shapes the architecture. It also shapes how we sell the platform. Operators who have had conversations with legal about AI procurement tend to find it relevant.

Where we are

Nexus AI is deployed and running in production utility environments today. It is not a roadmap. If you want to see it running against real data, the fastest way is a 30-minute call.