Sponsor session introducing Valency Bond — deep literature search over a fresh 45M-paper corpus backed by a 450M-edge citation graph, delivered as MCP tools inside Claude, Codex, or any agent — as the grounding data layer for open science in the agentic era.
Given in the Day 1 sponsor slot of the fully remote summer school; the video link starts partway into the day-long livestream, at the top of this segment.
“We really see ourselves as the data layer for this new era of agentic science. There's incredible intelligence that's happening above us, and when you ground intelligence in a real corpus, we think that is a huge power-up.” – Joshua Bloom
“Instead of having your LLM go off to the internet to do a web search, once you've connected up Valency Bond you'll wind up getting these answers very quickly, drawn not from what is on the web but from the actual record of what is in the corpus.” – Joshua Bloom
Auto-generated captions; may contain transcription errors. Josh's segment only (27:39–34:21 in the Day 1 livestream).
Thank you very much, Alex. I've got a couple of slides to show. Hello everyone — good morning, good afternoon, good evening, wherever you are. My name is Josh Bloom, and I am a professor at UC Berkeley and also CEO of Valency, and we're super happy to be supporting this summer school. I'll give you just a couple of slides here to introduce us. The big picture here is that we view AI-accelerated research as coming, and we want to support open science and open research in this era with a grounding. What I'll be talking about today is our first product, called Valency Bond, that we believe helps ground AI-accelerated research in a fresh corpus and helps against hallucinations — but, less on the negative side and more on the positive side, it can help you and your agents, and the ways in which you're working in this new era, do things with confidence and trust.
I should say that these slides are live — you can see the link down below, and there's a version of this talk up there if you want to follow along. But the biggest takeaway here is that we're hoping you can get a free account and start engaging with this large corpus of literature as you do your deep literature search using your favorite LLMs. So if you have a phone, you can take it out now and scan the QR code, or even better, go to the link I'm showing up there — app.valency.io/go/ns2026 — and you'll be able to connect your favorite LLM and get access to this fresh corpus of data.
Now, what can you do with that? Well, you can ask a lot of interesting questions, and do this with semantic meaning — so you're not just searching the fresh literature with keyword searches, but searching in natural language. If you want to know more about somebody that you're about to speak with, or who is coming up next — Vaishak's coming up after this — ask questions about that person and find out what they've been up to. And instead of having your LLM go off to the internet to do a web search, once you've connected up Valency Bond you'll wind up getting these answers very quickly, drawn not from what is on the web but from the actual record of what is in the corpus. So far we've indexed 45 million papers — there are a few thousand that come in every day from a variety of different sources — and behind that is an even larger graph of 450 million citations. So you wind up asking questions about the citation network: if we click on “show network,” you can see Vaishak's different collaborators across a couple of different areas that he's worked in. And that, we think, is incredibly powerful as you're trying to engage not just with the literature but with the people who are doing this work. And as you ask questions, and as you write your papers, you're interested in getting the artifacts that are important — like BibTeX exports of those papers of interest.
There are lots of different ways you can engage, across any topic of interest, but obviously in the neuro-symbolic world these are just a couple of examples: asking questions about trends, asking questions about impact across multiple fields, and again asking about people and their research and how it's had an impact on potentially other fields. If you go into the live presentation itself on the web, you can click on the “38 MCP tools” and see all the different tools that are happening under the hood, where we've done all the joins and the normalization of the data for you.
We also have a set of skills — with a single copy of what you see on the left-hand side into your terminal, you can install all these different skills, which are really just a collection of the MCP tool calls that allow you to go even deeper into the grounded corpus.
So, as I wrap up, just to say: we are in public beta right now. You still need an invite code, and for the whole summer school we've made one available — if you go to app.valency.io/go/ns2026, or you scan the QR code, which I'll show again, then you can just connect up your Claude, your Codex, whatever it is that you use in your daily life. And after you do that, you'll be able to confirm that you're live and start asking questions grounded in the corpus.
So with that, I'll say thank you. We're super excited to be supporting this summer school — please do connect. One of the fun things is, once you're in and you're working with your LLM, if you just say “send feedback to Valency” — if you see an error with the corpus, or you want something else that's not there — we'll see that, and we can act on it really quickly. The last thing I'll say is that we really see ourselves as the data layer for this new era of agentic science. There's incredible intelligence that's happening above us, and when you ground intelligence in a real corpus, we think that is a huge power-up, and we're super excited to see what you do with it. Please do get in touch, and have a great rest of the summer school. Bye, all.