AI Agents for Science

Building trustworthy, reusable AI agents with the scientific community—not just for it.

Artificial Intelligence (AI) Agents for Science is a collaborative community for scientists, researchers, developers, subject matter experts (SMEs), and others interested in responsibly applying AI agents to scientific work.

Supported by NASA's Accelerated Knowledge Discovery (AKD) initiative, this organization provides a shared space where community members can explore existing agents, exchange ideas, collaborate on scientific use cases, and create and publish agents that address real research needs.

Our Goals

We aim to:

• Develop practical AI agents that respond to real scientific needs.
• Enable community members to create, test, improve, and publish their own scientific agents.
• Encourage collaboration among scientists, SMEs, developers, and agent designers.
• Establish shared practices for trustworthy and responsible scientific agents.
• Promote transparent, reproducible, accessible, reusable, and accountable agent design.
• Keep scientific experts involved in reviewing assumptions, evidence, outputs, and decisions.
• Evaluate agents using realistic scientific use cases and benchmarks.

What You'll Find Here

This organization hosts community-developed AI agents and supporting resources for scientific and technical work. Available agents may help users:

• Discover scientific datasets and public software.
• Define project requirements and research needs.
• Organize evidence and identify potential research gaps.
• Explore NASA Earth observation imagery.
• Translate technical source material into traceable scientific illustrations.
• Support other research activities proposed and developed by the community.

Each agent should clearly describe its intended use, expected inputs and outputs, limitations, and areas requiring expert review.

Create and Share an Agent

Community members are encouraged to contribute agents that address meaningful scientific or technical needs. Scientists and SMEs can propose use cases and help define requirements, while developers and agent designers can contribute implementation expertise.

Contributors may:

• Develop and publish a new scientific agent within the organization.
• Collaborate with others on an agent idea or research workflow.
• Adapt reusable agent components for a different scientific domain.
• Contribute evaluation cases, datasets, benchmarks, or testing criteria.
• Improve an existing agent based on scientific or technical feedback.

Contributed agents should have a clearly defined purpose, documented limitations, appropriate safeguards, and sufficient information for scientists to understand and evaluate their outputs.

How Else Can You Participate?

You do not need to build an agent to contribute. Community members can also participate by joining the AI Agents for Science Slack Community to:

• Share scientific use cases and unmet workflow needs.
• Test agents using realistic research scenarios.
• Provide feedback on usability, accuracy, transparency, and reliability.
• Contribute examples, datasets, or evaluation criteria.
• Connect with potential scientific and technical collaborators.
• Share lessons learned from using agents in real scientific work.

To join the AI Agents for Science Slack Community, please email [email protected] for an invitation.

Our Approach

The agents shared here are intended to support—not replace—scientific expertise and judgment. Users remain responsible for reviewing outputs, verifying evidence, confirming scientific accuracy, and determining whether an agent or its results are appropriate for a particular purpose.

This organization is an open invitation to help shape the future of AI-enabled science while the practices surrounding scientific agents are still being developed.

Explore an agent, contribute a use case, or collaborate with the community to create and publish your own.