AI When the Wi-Fi Takes a Break
Local-first systems keep useful work close by and use the cloud where it adds value.
Body
The internet is excellent at many things. Being available every time you need it is not always one of them. What should an intelligent system still be able to do when the connection disappears?
Local-first design starts with that question. Keep the core data and useful functions nearby, then use cloud services for additional capacity or fresh information when they are available.
Keep a useful core
MMG’s local AI and infrastructure research treats the computer on site as part of the system, rather than simply a doorway to a remote service. Local storage can preserve observations and evidence; connected services can extend what the system does.
The mix depends on the task. A saved article can remain readable offline. Fresh external data still needs a connection. Local models also have hardware and capability limits, so moving a task onto a device does not automatically make it dependable.
Two experiments, different boundaries
NOMAD retrieves cited evidence from connected local archives without loading a model. The AIS pipeline stores vessel observations locally, but receiving new observations depends on its data feed. The public ship map currently shows a dated snapshot, not that continuously receiving collector.
Together, they help us ask which parts of a workflow can keep going and which need to wait. A useful system should explain that difference rather than treating a lost connection as a mystery.
Build for the interruption
Local-first does not mean giving up the cloud. It means choosing the dependencies deliberately and making a graceful fallback part of the design.
The aim is straightforward: preserve useful work when connectivity falters, and make it clear when an answer or observation needs an update. The Wi-Fi may take a break. The whole project should not have to.