For more than two decades, the Open Science community has worked to make research outputs open, findable, and reusable. Repositories, persistent identifiers, metadata standards, and shared infrastructures have built up an extraordinary base of publications, datasets, software, and project records.

The bigger challenge today is no longer access. It is making sense of the volume. Researchers rarely just look for a paper or a dataset. They need to connect evidence across outputs, disciplines, funders, and institutions, follow how ideas develop, and assess impact. Even with everything openly available, putting a reliable picture together still means moving between repositories, citation networks, project records, and institutional systems.

At the same time, AI is changing how research is discovered and used. A newer generation of systems, often called agentic AI, can reason across sources, plan tasks, follow relationships between entities, and support decisions based on structured evidence. For open scholarly infrastructures, this raises a clear question: can the systems that increasingly interpret scientific knowledge be grounded in open, transparent, community-governed sources rather than in opaque content of unknown origin?

A direct bridge between AI agents and the OpenAIRE Graph

To work on this, OpenAIRE has partnered with Alien Intelligence (Alien AI), a startup building agentic AI systems that interact with structured knowledge environments.

The collaboration combines two things. OpenAIRE brings the OpenAIRE Graph, a community-governed scholarly infrastructure with more than 350 million research products linked to researchers, organisations, funders, projects, datasets, software, and publications. Alien Intelligence brings the agent layer that can navigate these connections and turn them into usable answers.

The result is a direct connection between autonomous AI agents and the OpenAIRE Graph. Instead of relying on general web content or text prediction alone, agents can work with authoritative scholarly metadata, persistent identifiers, and verified relationships between research entities. Every answer can be traced back to its source in the Graph.

Paolo Manghi, OpenAIRE CTO, and Ghislain Delabie of Alien Intelligence describe the work as a commitment to trustworthy AI, addressing the two persistent issues in AI-driven research:

• Transparency. Each result carries a clear record of where in the Graph it came from.
• Quality. The Graph is rebuilt regularly, with deduplication and validation of ORCID and ROR identifiers, so the agents are not working from noisy or unverified data.

What this means in practice

The connection lets researchers, librarians, and analysts delegate multi-step tasks to an agent rather than working through them by hand:

• Literature reviews across a field, including gaps where new work could contribute.
• Citation analysis that goes beyond counts, showing how ideas branch and develop over time.
• Author and career mapping through ORCID, including collaborations and reuse of outputs.
• Dataset discovery across repositories, with an assessment of relevance to a specific question.
• Bibliometric signals on emerging topics, landmark studies, and high-impact work as it appears.
• Cross-domain links that surface methods and data from adjacent fields.

Why it matters

Until now, getting this level of insight from a large scholarly knowledge graph generally required data science skills, coding, and time. By placing AI agents on top of the OpenAIRE Graph, the same kind of analysis becomes available to a wider community: researchers, librarians, institutional leaders, funders, policymakers, and citizen scientists. Tasks that took days or weeks can be supported in a much shorter time.

Because the agents are grounded in the curated and continuously validated metadata of the Graph, the approach also helps protect scientific integrity at a moment when opaque AI systems and unverified content are becoming more common.

In a recent community demonstration, the Alien AI team showed how these agents can be installed as plugins in AI platforms. Another demonstration will be held during the OpenAIRE Graph Community Call on 20 May, 11:00-12:00 CEST which is open to the public and all are welcome to join. You can register here.

Looking ahead

As AI becomes part of everyday research, the infrastructures behind science become more strategic, not less. The question is no longer only whether research outputs are open. It is whether the systems acting on scientific knowledge are grounded in open, transparent, community-governed infrastructures. Bringing agentic AI to the OpenAIRE Graph is a step in that direction.