What if scientists could work alongside an AI collaborator that not only understands complex research goals but helps design and carry out experiments to achieve them?
That's the vision behind URSA — the Universal Research and Scientific Agent. The agentic open-source AI software package was developed at Los Alamos National Laboratory.
URSA is built to bring artificial intelligence into the heart of scientific discovery — acting as a team of specialized AI agents that can brainstorm hypotheses, plan experiments, run simulations and analyze results — all while learning and adapting along the way. In early demonstrations, URSA showed its potential in challenging domains like radiation-hydrodynamics, navigating intricate design spaces to find optimal solutions faster than ever before. By bridging human intuition with machine precision, URSA is a bold step toward a future where scientists and AI work side by side to accelerate innovation across the Lab.
Pushing the frontiers of fundamental AI
URSA is not just helping scientists — it is helping to redefine what artificial intelligence can do. At its core, URSA introduces a modular, feedback-driven agent architecture that breaks away from traditional linear AI workflows. Instead of moving through fixed stages of task execution, URSA's agents operate in dynamic, nested loops of reasoning, planning and verification, adjusting their strategies based on intermediate results. This adaptability allows URSA to handle the uncertainty and complexity inherent in real scientific inquiry.
Another key innovation is how URSA grounds large language models in the physical world. By integrating with real simulation tools and experimental data, the system does not just generate text — it reasons with equations, physical models and domain knowledge. This connection to real-world physics allows URSA to engage with simulations that reflect the actual laws of nature, creating a richer and more trustworthy form of machine reasoning.
To measure progress, the URSA team has developed a new benchmark framework to evaluate how well the system performs compared to established methods like Bayesian optimization. Early results show that URSA's "agentic" approach — where specialized AI agents collaborate and adapt — can lead to faster, more efficient and more accurate decision-making in complex scientific environments.
Addressing the Lab's most critical mission challenges
URSA's design aligns closely with the Lab's mission to tackle some of the most complex scientific challenges in the world. Many of these challenges — such as those in inertial confinement fusion, advanced materials or national security — depend on massive simulations and expert analysis that are both time-consuming and computationally expensive.
In early applications, URSA's prototype has demonstrated the ability to accelerate these workflows. For example, in ICF research, exploring the design space for optimal configurations traditionally requires running thousands of simulations — each one resource-intensive and costly. URSA streamlines this process by intelligently selecting and evaluating candidate designs, significantly reducing the number of runs needed to identify promising results.
Beyond fusion, URSA's agent-based framework could enhance simulation-guided discovery across multiple mission areas, from materials and manufacturing to stockpile modernization. By automating parts of hypothesis testing, data interpretation and optimization, URSA helps scientists extract deeper insights from supercomputing resources — not by replacing human expertise, but by amplifying it.
Looking ahead: Building the next generation of scientific AI
In the future, several avenues of research will further enhance URSA's capabilities. Ongoing research will focus on making the system more robust and reliable, particularly in managing errors or "hallucinations" that can occur when AI agents misinterpret data or miscommunicate with external tools. Improving these safeguards will be crucial for applying URSA in high-stakes environments where precision and trustworthiness are paramount.
Another key direction involves scaling URSA to operate across multiple domains simultaneously, enabling teams of AI agents to collaborate across chemistry, physics, materials science and beyond. The integration of human-in-the-loop interaction is also a top priority — allowing domain experts to guide, correct and refine URSA's reasoning in real time.
Together, these developments are bringing URSA closer to a new model of scientific discovery — one where humans and AI collaborate seamlessly to accelerate understanding, innovation and mission impact across the Lab.
LA-UR-26-22429
Contact
Media Relations | media_relations@lanl.gov
