AI Agents Now Run SETI’s Alien-Hunting Telescope: How It Works

The search for alien life just got an AI co-pilot. At the Allen Telescope Array in northern California, AI agents can now point the dishes, clean up the data and scan radio signals as they arrive, all from a single plain-English request.

No, they have not found aliens. But the first result is still a world first: a real-time AI detection of the Crab Nebula pulsar. Here is how the setup works, what hardware runs it, and how you can explore real SETI data on your own PC.

What Happened

On 30 October 2026, Nvidia published a blog post with the fun title “Things That Go Beep in the Night”. It describes how the SETI Institute and Breakthrough Listen now run agentic AI at the Allen Telescope Array (ATA).

A researcher types one natural-language request, and agents handle the chain of jobs a human team used to do by hand. The Decoder notes researchers “still type in their commands for now”, so humans stay in charge of what to observe.

Andrew Siemion, who leads SETI research at the SETI Institute and Breakthrough Listen, put it simply: “This isn’t something to fear. Rather, it’s a breakthrough in translation.”

Meet the Allen Telescope Array

The ATA was the first radio telescope designed from the ground up for SETI. It sits at the Hat Creek Radio Observatory, about 300 miles north of San Francisco.

How the AI Agents Work

According to Nvidia, one request can trigger the whole observing pipeline. The agents take over these jobs.

1. Steer the instrument. Point the array at the requested target.
2. Calibrate and focus. Tune the system before and during an observation.
3. Fix bad data. Catch and correct errors in the incoming stream, such as interference.
4. Run AI models live. Apply detection models to the data as it flows, not days later.
5. Plot results. Produce the charts a scientist needs to judge what was found.

The key idea is that inference runs on site, right next to the dishes. Radio data is too big to ship to a cloud data centre in real time.

Note: Nvidia does not say which language model drives the agents. It says the team is evaluating its open Nemotron models for cases where keeping data local matters most.

The Full Tech Stack

If the RTX PRO 6000 is new to you, my workstation GPU guide explains where it sits against consumer cards.

The Crab Pulsar “First”

The team reports the first real-time AI detection of a pulsar in the Crab Nebula. A pulsar is a spinning neutron star that sweeps a radio beam past Earth like a lighthouse.

Why does a known pulsar matter? Because it is a perfect test signal. If the agents can spot a real, repeating cosmic signal live, the same pipeline can flag something unusual the moment it appears.

Nvidia says the setup compresses work that would take humans weeks, or even years, into hours. These are vendor and team claims; no independent paper on the agent system has been published yet.

From Last Year’s FRB Breakthrough to Agents

This did not come out of nowhere. In November 2025, the SETI Institute and Breakthrough Listen reported a Holoscan-based AI system for spotting fast radio bursts (FRBs) at the ATA.

The old pipeline needed about 59 seconds to process 16.3 seconds of data, nearly four times slower than real time. That gap is what the AI work set out to close.

How the ATA Compares With Other SETI Telescopes

Explore Real SETI Data on Your PC

You cannot run the ATA’s agents at home, but you can look at real SETI recordings. Breakthrough Listen publishes data from Green Bank and Parkes for free.

What you need

• Python 3.9+ on Windows, macOS or Linux.
• Disk space: single files can be several gigabytes, so start with one.
• RAM: 16 GB is comfortable for small files; see my RAM guide for bigger jobs.
• GPU: optional. The tools below run on CPU.

Step by step

1. Read the primer. The Berkeley SETI Research Center’s Listen pages explain the file types (filterbank .fil and .h5).
2. Find a file. Open the Breakthrough Listen Open Data Archive at breakthroughinitiatives.org/opendatasearch. Filter by project (Green Bank or Parkes), file type (HDF5 or filterbank) and target name.
3. Create a Python environment. Run python -m venv seti, then activate it.
4. Install the tools. Run pip install blimpy turbo_seti. Both come from Berkeley SETI.
5. Inspect the file. In Python: from blimpy import Waterfall, then wf = Waterfall("file.h5") and wf.info().
6. Plot it. Run wf.plot_waterfall() with matplotlib to see frequency against time.
7. Search for drifting signals. Run turboSETI file.h5 -M 4. It looks for narrowband signals whose frequency drifts, the classic “technosignature” pattern.

Tip: Almost every hit you find will be radio interference from satellites, phones or Wi-Fi. That is normal, and filtering it out is exactly the job the ATA’s agents now do automatically.

Troubleshooting the SETI Tools

• MemoryError when loading a file: load a slice instead with Waterfall("file.h5", f_start=..., f_stop=...) using frequencies in MHz.
• HDF5 or hdf5plugin errors: reinstall with pip install --upgrade h5py hdf5plugin; many files use bitshuffle compression.
• No plot window appears: you are probably in a terminal without a display. Save instead with plt.savefig("plot.png").
• turboSETI finds thousands of hits: raise the signal-to-noise threshold with -s 25 and compare “on” and “off” target files, as Breakthrough Listen does.
• Download is huge or slow: pick an .h5 product rather than baseband data, which is raw voltage and far bigger.

Limits, Hype and What Is Unknown

• No alien signal: SETI has searched since 1960 without a confirmed detection, and this system has not changed that.
• Vendor claims: the speed and “first” claims come from Nvidia and the team, not a peer-reviewed paper on the agent system.
• Which LLM: not disclosed. Nemotron is only “under evaluation”.
• Code availability: CyberEther is open source, but Nvidia does not say whether Nexus or the agent setup will be released.
• Human oversight: researchers still issue the commands; the agents do not choose targets on their own.

Note: The team’s long-term idea is offline observatories on the far side of the Moon, where there is no radio noise from Earth. That is a research goal, not a funded mission.

My Take

I love this story because it is AI agents doing something genuinely useful and a bit magical. The bottleneck in SETI was never just the dishes; it was the mountain of noisy data that no human team could sift in real time.

The same pattern, agents plus local GPUs next to the data, is what many of us run at home with Ollama or ComfyUI on a smaller scale. If you want to try the “local AI next to your data” idea yourself, my best Ollama models guide is a good start.

Will the agents find ET? Probably not soon. But they make the search faster and cheaper, which raises the odds over time.

FAQ

Did AI find aliens at SETI?

No. The AI agents detected a known pulsar in the Crab Nebula in real time. No alien signal has been found.

What do the AI agents do at the Allen Telescope Array?

They steer and calibrate the telescope, fix data errors, run AI models on the live stream and plot results, all from one natural-language request.

What hardware does SETI use for this?

Nvidia RTX PRO 6000 Blackwell GPUs running Nvidia Holoscan, on site at the observatory, handling up to 2.4 terabits per second.

Which AI model controls the telescope?

Nvidia has not named it. The team says it is evaluating open Nemotron models for local use.

Can I help search for aliens from home?

SETI@home stopped sending new work in 2020, but you can download Breakthrough Listen data and analyse it with the free blimpy and turboSETI tools.

How many dishes does the Allen Telescope Array have?

42 dishes, each about 6 metres across, at Hat Creek in northern California.

Is the SETI agent software open source?

The CyberEther signal-processing framework is open source. Nvidia has not said whether the Nexus orchestrator or agent setup will be released.

Sources

• Nvidia blog: Agentic AI joins the search for extraterrestrial signals
• The Decoder: AI agents take over at SETI’s Allen Telescope Array
• SETI Institute: Allen Telescope Array
• SETI Institute: 600x AI speed-up for FRBs (2025)
• Breakthrough Listen Open Data Archive