SETI Institute adds AI agents to steer a telescope in the search for alien signals

AI agents can now help steer the Allen Telescope Array in the search for extraterrestrial signals, but the reported demonstration is a pulsar. NVIDIA said October 30 that Breakthrough Listen and the SETI Institute have connected agents to live telescope operations, letting a single natural-language request trigger instrument control, data correction and AI analysis.

From a request to an observation

At the Northern California observatory, the integration gives agents a role beyond examining a finished observation. They can point the telescope’s antennas, calibrate and focus the instrument, correct data errors and plot the result. The workflow links control of the physical instrument with software that processes its incoming measurements, rather than treating those as separate tasks.

The coordinating layer is Nexus, a web-based observatory manager. It acts as the agent harness: the software that connects an agent to the information and tools it needs to act. Nexus provides a shared view of the telescope’s state, available computing resources and processing pipelines, so those pieces can be managed together.

Researchers can also use natural-language prompts to inspect and change the observation workflow through Nexus. NVIDIA describes three concrete capabilities:

• Monitor packet loss, checking whether pieces of the incoming data stream are being lost.
• Read live spectrograms, visual displays of the signals arriving during an observation.
• Rewrite signal-processing blocks, changing parts of the software workflow that handles those measurements.

Processing the stream before it fills drives

Underneath that interface, NVIDIA’s Holoscan platform handles the live sensor stream. A Holoscan library called DAQIRI moves telescope data directly into graphics-processor memory. That allows observations to be processed as they arrive, rather than first accumulating on hard drives. It also supplies the live data on which the agents act.

CyberEther, a signal-processing framework developed at the SETI Institute, performs the mathematical work. It uses NVIDIA’s CUDA-X software libraries, including cuFFT and cuBLAS. The division of labor is specific: DAQIRI brings measurements into GPU memory, CyberEther processes the signals, and Nexus coordinates the instrument, computing resources and software workflows.

2.4 terabits per secondReported data-handling capacity

NVIDIA says the system handles this stream on RTX PRO 6000 Blackwell GPUs, with AI inference running on site beside the telescope.

A pulsar milestone, not an alien signal

The reported scientific demonstration is a detection in the Crab Nebula. NVIDIA describes it as the first real-time AI detection of a pulsar—a rapidly spinning stellar remnant that emits repeating pulses. That is the claimed milestone for the pipeline, distinct from its mission of searching for extraterrestrial intelligence. The first-of-its-kind designation remains NVIDIA’s claim.

The model choices are still being explored. The team is evaluating open models, including NVIDIA Nemotron, for situations where keeping data local and getting results quickly are priorities. That is an evaluation, not a statement that Nemotron powered the reported pulsar detection. The live pipeline and the next model-selection work are different stages of the deployment.

NVIDIA also connects that evaluation to a more distant possibility: observatories that might someday operate offline on the far side of the moon. That remains a prospective use case. The concrete deployment described now is at the Allen Telescope Array, where agents have access to both the instrument and its incoming data.