AI-assisted science with rigor at its core
Lightcone Research is an international initiative building infrastructure for rigorous AI-assisted computational science. Our open-source tools and schemas create a standardized research substrate that captures the scientific decisions, evidence, and provenance behind a result, giving each analysis a structured, inspectable record that establishes soundness and mechanically verifies provenance. Our tooling is designed as a sidecar to be maintained by AI agents as the analysis evolves, providing automatic structuring and formalization without any additional overhead for the scientists driving the research.
AI is changing what’s tractable in scientific research, enhancing scientists' ability to tackle ambitious and far-reaching research questions; however, it also risks outpacing the community’s already strained capacity to scrutinize new work, thereby eroding the rigor that makes the scientific cannon worth building on. Without infrastructure that makes AI-assisted workflows legible, the production of new work risks outpacing our ability to evaluate it, eroding the rigor that makes the scientific record worth building on. Lightcone Research aims to ensure that the expanded reach enabled by AI strengthens that record rather than weakening it. For a more detailed version of our vision, read our perspective paper.
Guided by the vision for BIDS as a space for open scholarship, open source software, and interdisciplinary collaboration on AI in science and society, the Lightcone Research initiative and its community contributes to a shared common goal of advancing scientific inquiry and technological development.
Projects
Agentic Schema for Transparent Research Analysis (ASTRA)
Agentic Schema for Transparent Research Analysis is a YAML specification for scientific analyses that provides a structured, inspectable record for a given workflow. This record lives between the code and the paper, and captures the inputs, choices, and reasoning behind the produced results, making it easy to quickly establish soundness as well as to verify provenance. ASTRA is designed to integrate with any agentic harness and add very little overhead, as the agent maintains the record itself, rather than asking the researcher to formalize it by hand.
lightcone-cli
The lightcone CLI and associated skills support the development of scientific analyses with ASTRA. This includes workflow management systems, designed to ensure provenance and mechanical reproducibility, as well as high-performance computing support, workflow packaging, literature search, and more.
Community Events
We believe that the future of Agentic AI for science will take the form of a rich ecosystem of interoperable tooling covering all the stages of scientific research, from ideation to publication. To that end, we are committed to building in the open, as well as to engaging with the community to facilitate the emergence of this ecosystem and related standards. Most recently, we hosted the Agentic AI for Science Developer Summit during Summer 2026 - stay tuned for more events.
Partners and Supporting Organizations
Lightcone Research is an international initiative based out of the Berkeley Institute for Data Science at UC Berkeley and the AI for Science and Science for AI Center at the French National Centre for Scientific Research (CNRS), with support from Eric and Wendy Schmidt.
Join Us
Get involved with Lightcone Research through project contributions and at events!
To stay in touch, please join the Lightcone Research mailing list by visiting the Google Group page or emailing bids-lightconeresearch+subscribe@lists.berkeley.edu, and follow us on LinkedIn. If you want to learn about shaping the community-owned ASTRA standard, visit the ASTRA website.
To hear about all BIDS events and news, follow us on Bluesky & LinkedIn and subscribe to the BIDS newsletter.
News and Events
- Event: Agentic AI for Science (AAI4Science) Developer Summit • July 28-31, 2026 • AI Futures Lab, Berkeley, CA
- Publication: Science that Compounds: The Need for A New Substrate for Research in the Age of AI • Lanusse, F., Parker, L. • May 13, 2026
