AI inference
A trained model can process data close to the application. Research focuses on selecting suitable hardware and measuring latency.
Edge computing
JetGrid is developing a network of physical nodes coordinated by shared software. The aim is to run workloads close to where data is generated, from a building sensor to a local service.

The platform
Where a workload runs matters. JetGrid is studying how to select a node according to an application’s resource needs and the location of its data.
The physical unit provides processing power, memory and graphics acceleration at the installation site.
Software collects information on node health. Assigning workloads according to resources and availability is the next step.
The service defines what to process and what result to return, such as an event detected in a video stream.
Tested prototype
On 24 March 2026 the JetGrid project verified registration, telemetry and recovery after reboot on an NVIDIA Jetson AGX Orin. A second host confirmed that registration could be repeated.
Read about the testApplication areas
A trained model can process data close to the application. Research focuses on selecting suitable hardware and measuring latency.
The proposed use case processes a stream locally and transmits selected events. Detection criteria need to be validated for each deployment.
Museum applications could associate multilingual audio and other content with stops on a visitor route, including locations with intermittent connectivity.
Company disclosures
| Legal name | JetGrid S.r.l. |
|---|---|
| Registered office | Viale Bruno Buozzi 3, 00197 Roma, Italy |
Our registered office in Rome handles institutional correspondence and enquiries about collaboration.