AI agents join live radio-telescope operations at the Allen Telescope Array AI agents join live radio-telescope operations at the Allen Telescope Array

NVIDIA says the system processes incoming observations near the telescope, using Holoscan data ingestion and local GPU inference. Nexus supplies the orchestration layer. The company reports that agents can control the instrument and work with incoming measurements through a language interface. It also reports a real-time AI detection of a pulsar in the Crab Nebula.

The Stelline project documents the infrastructure beneath that interface. Its stack separates packet ingestion, signal processing, storage and cluster orchestration. CyberEther handles signal processing and visualization, while the Stelline plugin provides telescope networking and radio astronomy tools. DAQIRI covers packet ingestion, and Nexus coordinates distributed execution of observations across the compute cluster.

Those components show where an agent fits into an observatory. Scientific software still has to ingest the stream, transform the measurements and retain the relevant results. The language interface gives operators another way to work with that machinery. It depends on the instrument and processing layers continuing to expose usable state and reliable controls.

For teams building other sensor-heavy systems, the deployment offers a specific architecture to inspect: keep inference close to incoming data and connect it to explicit processing tools. The documented separation of networking, computation, storage and orchestration makes those responsibilities visible. It also gives developers places to test the system before adding authority over physical equipment.