The United States is a leader in nuclear power research, yet a potential treasure trove of new breakthroughs and insights lay sequestered in archives of millions of reports, engineering drawings, operational logs, and photographs. Currently, before a document can be released out in the open, a trained expert must confirm that it is safe to share. There are few people qualified to make that call, and they have to work through the archive one document at a time. This bottleneck limits access to valuable research and historic innovations, leading engineers to repeat work to rediscover what may already be known.

Today, we get to help change that, and we are very excited about it.

HERALD, the project we are building with senior scientists Daniela Ushizima, Peter Nugent, and colleagues at Berkeley Lab, has been selected for a Department of Energy award through the Genesis Mission.

The project will use special-purpose AI agents leveraging Valency’s infrastructure to help experts review this enormous archive.

View larger

×

Our Valency team is proud to support the Genesis Mission and the acceleration of American science innovation.

HERALD stands for High-throughput ECI Review and Agentic Legacy Document processing. It is exactly the kind of ambitious, consequential problem we started Valency to work on.

The official announcement is here: https://genesis.energy.gov/

Launched by executive order in November 2025, the Genesis Mission brings the Department of Energy’s seventeen national laboratories together with industry and academia. Its goal is to double the productivity and impact of American science and engineering within a decade.

Our selection came through a highly competitive process, and Valency is one of only a few dozen private companies to be funded through Genesis. Earning national recognition like this at an early-stage startup makes the moment especially rewarding.

What HERALD does

HERALD gives AI agents the first pass through the archive. They will read text and images together, handle routine material, flag anything that needs closer attention, and show the evidence behind every recommendation.

The project will begin with data that is already public. This gives the team a safe foundation for developing, testing, and evaluating the system.

The experts remain in charge of every decision that matters. The difference is that they can spend their time on the documents that genuinely need their judgment.

Where Valency comes in

Most document systems were built to store files and help people find them. They were not built for AI agents that need to read, reason across, and act on huge collections of documents alongside human experts.

We think of this as research intelligence: turning vast scientific collections into evidence that people and AI agents can reason over together.

That is the system Valency has been building. For HERALD, it means supporting hundreds of millions of scientific documents, many AI agents working at once, and a large group of human reviewers, all inside the tight security boundaries this work demands. Valency also speaks the Model Context Protocol natively, giving HERALD’s agents a direct, structured way to work with the archive.

HERALD is the first real-world use of this infrastructure. We could not have asked for a more meaningful place to start.

The archives are not going to read themselves, and they are only the beginning. If helping solve big scientific problems at very large scale sounds like your kind of fun, we are hiring!. Or just say hello 👋 team@valency.io.

If you haven’t seen Anthropic’s announcement of Claude Science, their new workbench for enabling the scientific workflow, it’s well worth a read. They’re the latest entrant into a fast-growing ecosystem of AI tools built to accelerate the pushing of boundaries. Google has its Co-Scientist (now peer-reviewed in Nature), OpenAI has Prism, and now Anthropic has Claude Science.

But the AI-accelerated research ecosystem isn’t just made up of products from the frontier model companies. Smaller, independent labs are building real systems too — like FutureHouse’s Robin, a multi-agent system for experimental biology. The ecosystem is growing in other places, too: AI authoring systems, AI reviewing systems, co-scientist systems, and a host of other providers. At Valency, we’ve been thinking a lot about this ecosystem since our inception and about how all of these components might interlock in the agentic science era.

In some ways, our initial offering Valency Bond shares some of the same components as Claude Science. We make reasoning agents smarter at doing science by giving them easy access to real scientific articles. We even surface some of the same data sources. But ultimately, we think Valency Bond is a uniquely positioned offering in this ecosystem. Here’s why:

1. Built for academics and non-profits. Valency Bond is people-centric, designed first and foremost around the needs of researchers at universities and non-profit institutions. We don’t charge for their access.

2. LLM-agnostic. Whether you’re using Gemini, ChatGPT, or Claude, our connector plugs into your scientific and research workflows. And through bearer tokens, you can use Valency in your own LLM harness — including with open-weight models in your own front end.

3. Cross-source data quality. This one is more subtle, and perhaps more important. Any single data source served through MCP comes with its own data-quality challenges. As we’ve brought more and more sources into our growing corpus, we’ve found that intercomparing across all of these different sources is a powerful way to clean them up — and to provide better, fresher answers to our growing community. On top of that, our toolkit includes capabilities like semantic search and trending topics. We’re distilling the actual questions researchers are asking into tools that give them better, faster answers.

4. Discovery and ideation happen across fields. Serving individual data sources as separate tools risks the siloing effects we as scientists want to avoid during discovery and ideation phases. De-balkanizing search across academic and other data sources is extremely potent and we’ve taken steps to ensure that is how our users can work.

Today’s announcement is a great reminder that this new agentic science ecosystem is just starting to flourish — and we welcome Anthropic’s entrance. This AI-accelerated scientific ecosystem will only broaden, and we can’t wait to tell you about our new product offerings. Watch this space — and definitely sign up to use Valency Bond.

Now, as academics are starting to use LLMs not only to accelerate the ideation phases of research but for the paper writing process itself, a new avenue of trouble is arising en masse and at scale: the COmmission of fabricated citations. When people do this, to support a claim in their work e.g., it’s considered fraud. When LLMs do it, we call it hallucinations. The work of Topaz et al. published last week (Fig. 1) makes clear that fabricated citations have been steadily increasing since frontier models started gaining more adoption. In March, Adam Ruben explored and explained some of the origins of why LLMs might hallucinate and why it’s challenging for us humans to catch all such errors.

LLM-hallucinated citations have been a concern by those paying attention for years. And like citation omission, the commission of LLM-hallucinated citations were likely just viewed as a new occupational hazard, becoming yet another part of the imperfect system.

Then, yesterday, Thomas Dietterich a moderator and principal at the premier preprint server arXiv made a startling announcement: hallucinated citations would not only not be tolerated, if an author is caught submitting a paper with a hallucinated citation they will be banned for 1 year from submitting to arXiv. The existence of hallucinations, reckons arXiv, is a red-flag indicator that a human author did not actually write, read, and/or verify the context they signed their name to. As submissions to arXiv are a critical venue for academics to present their work, such a banishment would severely hamper an academic’s ability to perform a basic facet of our work. Letting an LLM hallucinate on your behalf can now lead to a serious and consequential blow to your reputation and, by extension, your livelihood. Hallucinations aren’t just an academic annoyance, they are a new threat vector.

Hallucinations are one of the problems in the AI/science space I’ve been sitting with and concerned about for the past year. So, I wanted to build something to help. Last week my company announced Valency Bond, in limited preview, to give LLMs access to a real corpus of 50 million+ academic papers and preprints. While the use cases for Valency Bond are wide, we’ve found that frontier LLMs connected to Valency Bond become more tethered to the relevant work and those works are surfaced as real citations to the researcher interacting with the LLM. We still need to sign off on the work we submit but my hope is that products like Valency Bond will help tamp down this new threat vector and unlock our ability as academics to accelerate our research with AI.

References

1. Adam Ruben “Cite unseen: when AI hallucinates scientific articles” (Science, 0.1126/science.zfae4z1; 30 Mar 2026)

2. Maxim Topaz, Nir Roguin, Pallavi Gupta, Zhihong Zhang, Laura-Maria Peltonen “Fabricated citations: an audit across 2.5 million biomedical papers” (Lancet Volume 407, Issue 10541 p1779-1781; 9 May 2026)

I’m excited to introduce you to Valency. We’re building a home for people-centered, AI-accelerated research. Today we’re giving a sneak peek to Valency Bond – a new model context protocol (MCP) service that connects frontier AI models to a structured, embeddings-indexed corpus of the scientific record.

Why Valency Bond?

As scientists, we seek out every new opportunity to advance our fields, using state-of-the-art instruments, starting new collaborations, and leveraging new technologies. Galileo was scrappy in this way, famously co-opting the telescope (invented for military purposes) to study the heavens. In doing so, he ushered in the era of modern astronomy. As an astronomer, I think a lot about Galileo’s precedent as a deft and talented tool user.

Now, given the undeniable potency of AI, it’s natural for modern scientists to look to frontier large language models (LLMs) as the singularly most transformative technology of our lives. But while the LLMs many of us use in our personal lives are impressive, they haven’t matched the quality and depth needed professionally. Most researchers have caught (or not!) an AI confidently citing a paper that doesn’t exist. Moreover, AI models were trained on yesterday’s science—they don’t know what was published last week, last month, or even last year. When asked about recent or specialized work, they can hallucinate. And AI-generated content is already outpacing our ability to verify it.

Valency Bond is our answer to that. Think of it as the layer that helps keep the AI working from the real scientific record, not a plausible-sounding approximation of it. More viscerally you could think of it as the data artery that feeds the (LLM) brain. Valency Bond surfaces tens of millions of papers and preprints and grows daily, with coverage from 1965 to yesterday.

With 37 MCP tools, LLMs connected to Valency Bond are a powerful new venue for ideation, grounded in the real literature. With Valency, ask your favorite LLM for a few new research directions you might take or to propose some new collaborators for you on a topic you’re interested in. When I travel to other institutions, I’ve started asking Valency for help in finding deeper academic connections to the people on my meeting schedule and to identify those in other departments who might have intellectual overlap.

Early researchers in academia, government labs, and industry have been kicking the tires for a few months and getting some of the wow! moments that we had hoped for. Dr. Peter Nugent, Senior Scientist and the Division Deputy for Science in the Applied Math and Computational Research Division at Lawrence Berkeley National Laboratory (LBNL), said “Valency has transformed how I approach literature discovery, compressing what used to take days of manual searching into a few targeted queries taking minutes. Having cross-corpus semantic search and citation graph tools in a single interface has meaningfully accelerated the pace at which I can move from a research question to a well-grounded reading list to scientific discovery.”

Why Valency?

This is a particularly interesting, if not singular, moment for researchers as AI permeates more and more of the scientific enterprise. We’re all navigating these potent changes afoot with varying levels of excitement and trepidation. My co-founder Ryan and I started Valency recognizing that the new era of AI-accelerated science is still, and will be, fundamentally, a human-centered endeavor. And as we all step into this world the tools and frameworks we use should allow us (and our AI agents) to continue to build on the shoulders of giants, with confidence.

Watch this space in the coming weeks to learn more about what we’re building, how we’re building, and who we are.

—

If you’d like to try Valency Bond, please join our waitlist. I’m so curious to know what you’ll find—please get in touch.

-Josh
Valency CEO, Co-Founder