Agents steer Allen Telescope Array and make first real-time AI pulsar detection

Agents built on NVIDIA hardware can now steer the Allen Telescope Array, correct the data it collects and apply AI models, all through natural-language requests. The system has already made the telescope's first real-time AI detection of a pulsar. The pipeline handles 2.4 terabits per second, which is how much sensor data passes through it while observations are still running.

Volume is the central problem. Radio telescopes produce enormous amounts of data, and scientists must pick out potentially interesting signals while observations unfold. According to NVIDIA's blog post on the work, DAQIRI, a library in the Holoscan platform, pulls the telescope's sensor data directly into GPU memory so it is processed in flight instead of filling hard drives. CyberEther, a signal-processing framework developed at the SETI Institute, performs the calculations using NVIDIA CUDA-X libraries including cuFFT and cuBLAS. The processing runs on NVIDIA RTX PRO 6000 Blackwell GPUs, and inference happens on premises, next to the instrument.

An agent layer sits on top of that stack. Nexus, a web-based observatory orchestrator, gives agents a single view of instrument state, compute resources and pipelines. Researchers can watch packet loss, read live spectrograms and rewrite signal-processing blocks by prompting in natural language. According to the blog post, the combined system lets an agent steer the telescope's antennas, calibrate and focus the instrument, catch and fix data errors and plot the result. NVIDIA claims this compresses work that would take humans weeks or years into hours. Those are NVIDIA's claims, and the post gives no benchmark for the comparison.

The blog post describes it as the first real-time AI detection of a pulsar, a rapidly spinning stellar remnant, in the Crab Nebula. The post does not say how many detections the system has made, how it performed against conventional pipelines, or whether the detection was previously known. The team is also evaluating open models, including NVIDIA Nemotron, for cases where data locality and turnaround matter most, and for observatories that might someday run offline on the far side of the moon.

Breakthrough Listen and the SETI Institute are bringing agentic AI into the daily search for extraterrestrial intelligence. Andrew Siemion, Bernard M. Oliver chair for SETI at the SETI Institute and principal investigator of Breakthrough Listen, said that as these agentic observatories reach out, "what they find may have much more in common with themselves than with the humans who created them," adding that "this isn't something to fear." The search itself changes how a radio observatory is run. An operator who describes a goal in plain language sets the instrument and the analysis in motion, and the software decides how to act on the live data stream.