Delivery to the edge is getting solved
Fleet management for tactical devices had a busy spring. The latest entrant, UDS Fleet, launched June 2 with an Android app and a browser client for deploying and managing software across distributed tactical systems. Programs that once hand-carried every update now have real options for pushing signed software to devices they cannot reach over a network.
That is hard engineering. Air-gapped networks cannot pull from container registries. Classified environments require a custody record for every binary. Disconnected devices need store-and-forward updates. The teams solving delivery are solving a problem the rest of the industry ignored for a decade.
Delivery is infrastructure. Operations are what the operator cares about.
A fleet tool gets a container, a model file, or an application onto a tactical device reliably and with provenance. That is the delivery layer. It does not decide what the software does once it is running. Local AI, mission-aware routing, sensor fusion, and evidence capture all sit above it.
Getting a container onto a constrained platform is impressive infrastructure work. Running local threat assessment on that platform's sensor feeds without reachback is a different kind of software entirely.
Two layers, nobody owns both
Layer one gets software to the device: containerized delivery, air-gap support, signing, provenance, and fleet-wide updates.
Layer two makes the software fight. Local inference on the device hardware. Mesh networking when the cloud is unreachable. Sensor fusion across cameras, microphones, and RF receivers. Mission state that persists through reboots and link drops. Evidence capture with chain of custody, and missions that wipe clean when they end.
Neither layer works without the other. The best operational software is useless if it cannot reach a classified device. The best delivery pipeline is useless if the software it carries cannot operate on its own at the edge.
What EdgeLance runs after delivery
An EdgeLance mission pack arrives on a device as a signed bundle: model weights, mission configuration, mesh credentials, and evidence schema. The capacity policy engine reads the device's hardware profile and loads the models it can carry. The mesh router opens whatever links exist. The evidence store initializes its encrypted containers.
From there the device is an autonomous tactical node. It runs inference on camera feeds, fuses sensor data, publishes entities over Cursor-on-Target for TAK, and records every detection with its provenance. None of it needs a connection to a cloud or an operations center.
When the mission ends, the operator decides what to promote upward and what to destroy. The delivery layer never needs to know. It delivered the software. EdgeLance made it fight, and a laptop command node fed by any air-gapped delivery pipeline is a complete stack from repository to mission completion.