£50k in prizes. Nov 15-16. Be there. Holistic AI, UCL Centre for Digital Innovation (CDI), Amazon Web Services (AWS), Valyu.
About us
Valyu is the information layer for AI systems. We connect agents and LLMs to structured knowledge across the web, finance, research, medicine, and more through a single unified Search API. Trusted by leading teams to power RAG, agent workflows, and knowledge applications.
- Website
-
https://www.valyu.ai/
External link for Valyu
- Industry
- Technology, Information and Internet
- Company size
- 2-10 employees
- Headquarters
- San Francisco
- Type
- Privately Held
- Founded
- 2023
- Specialties
- Machine Learning, LLMs, RAG, Attribution, AI Search, and Context Enrichment
Locations
-
Primary
Get directions
San Francisco, US
-
Get directions
London, England, GB
Employees at Valyu
Updates
-
Valyu reposted this
If you’re building AI agents or research workflows in healthcare, you know how painful it is to access clinical trial data. The ClinicalTrials.gov API is powerful, but built for humans, not AI. Essie syntax, MeSH terms, pagination tokens, nested JSON… it’s unusable for agents. The 𝗩𝗮𝗹𝘆𝘂 𝗦𝗲𝗮𝗿𝗰𝗵 𝗔𝗣𝗜 fixed that. Our Clinical Trials Search is the first natural language search interface over 470,000+ trials. No more complex tool calls, agents just ask: “Clinical trials in the UK for COVID-19 patients evaluating low-dose dexamethasone.” A must-have for anyone building AI in healthcare 👇
-
Rustaceans we have a Rust SDK 🦀 Start building at the speed of now (ATSON) 🛠️🫡
🦀 We care deeply about our users. When something doesn’t work for them, it hits a nerve and we make it right at the speed of now. This afternoon, a user messaged me around half three. 10 mins later we joined on a call and learnt that he was trying to integrate the API in his Rust project, his experience would have been so much better if we just had a Rust SDK. By 5.30, we shipped one- alpha version, comprehensive docs, clean examples, and some advanced features, all with idiomatic Rust. It has become second nature for Alexander Ng, Harvey Yorke, and I; when something doesn’t work for our users, it genuinely gets under our skin and we try do everything we can to fix it for them atson (At the Speed of Now). Rustaceans check out our shiny new Valyu AI Search API Rust SDK 🦀 https://lnkd.in/e4UVeprg (would love feedback).
-
Valyu reposted this
Unlike other search APIs, the Valyu Search API has access to live structured financial data. This is our API retrieving insider activity from Palantir, with no api parameters other than a natural language query need. What does this mean for agents? They can now finally access the financial data they need for the most difficult financial deep research workflows 👇
-
It's official... The Valyu Search API is the most powerful search API for financial AI agents and deep-research applications.
We set out to build the toughest financial search benchmark we could imagine. Why? Because the real world of finance research is far more complex than any existing benchmark suggests. Analysts don't ask simple questions; they scour noisy SEC filings, analyse fast-breaking news, and connect dots across disparate sources. We knew that for AI to be truly useful here, it needed to perform under those conditions. So, we put Valyu's Financal Search to the ultimate test against Google, Exa, and Parallel. Most search systems and benchmarks fail there. They assume they are getting back clean and fresh results and skip multi-step workflows. So we built a benchmark that reflects how real financial research works. Our Finance Benchmark consists of four parts: SEC Filings: Dense, noisy, structured. Financial News: Freshness + relevance under time pressure. FinLitQA: Core literacy from textbooks and references. Multi-turn QA: Evidence integration across queries and sources. We ran all four APIs through the same, simple tool-call integration into an AI agent. Results: Valyu 73%. Google 55%. Exa 63%. Parallel 67%. What the numbers don't show is how other systems failed. Some missed critical 10-Ks. Others surfaced outdated news on time-sensitive tickers. Why this matters: financial AI breaks when your data is unreliable. Data for finance is notoriously fragmented across legacy APIs, inconsistent schemas, and stale indexes. If your AI agents cannot reliably ground their answers in up-to-the-minute market data, filings, policy documents, and comprehensive financial literature, then what they are doing isn't research, it's merely an educated guess. Valyu was built for this. One unified search layer for built for agents: • Fresh financial news as it breaks • Comprehensive SEC filings • Economic data structured for your AI • General web search when you need it If your AI still relies on search that cant meet the demands of real workflows, fix the pipeline. Integrate Valyu.
-
-
Valyu reposted this
“Why can’t we just use existing traditional search APIs when building AI applications?” Because a biomedical research assistant demands the content of clinical studies, not a link and snippet from an seo blog… We sat down with the legend Harnoor Singh recently on his youtube channel to discuss this at length (link below) - it was amazing to show what the London startup/research ecosystem has to offer 🙌 (and why most move to sf...) At Valyu we built a search experience from the ground up for AI agents. They are the primary consumer, and will soon greatly outnumber us in their consumption of information. We built the worlds best search API to serve them. In our recently released benchmarks we showed just how powerful the API is, with SoTA benchmarks across SimpleQA and FreshQA, as well as vertical specific benchmarks in finance/medicine/economics that test how well agents with our search API perform on real-world tasks. Here are the results from our latest financial benchmark: Valyu (73%) Google (55%) Exa (63%) Parallel (67%)
-
We benchmarked our Search API in realtime retrieval against Google, Exa, and Parallel. We came out on top. When we started building our Search API, we knew real-time performance was a requirement for AI adoption across knowledge work domains and workflows. Other Search APIs depend on stale indexes and delayed recrawls that fail the moment an agent needs to reason over fresh inputs. FreshQA was the first benchmark we used to test that assumption. FreshQA is a rolling dataset of 600 time-sensitive queries updated weekly. Some questions are about ongoing geopolitical events. Others track market volatility, legal developments, or cultural news. What matters is that they change, fast, and that answering them requires retrieval systems to reflect the current state of the world. We ran FreshQA across four major APIs. Each was evaluated using the same integration method: a simple tool call, passed into an LLM (Google, Exa, Parallel), and judged using strict accuracy labels: correct, partial, or incorrect. Valyu scored 79%. Parallel 52%. Google 39%. Exa 24%. What the numbers don’t show is how systems failed. Some returned cached results, others couldn’t handle time sensitive queries, and one API often surfaced links from months ago - this is useless for AI agents. Crucially, Valyu isn’t just “fresh news.” With us your agents can retrieve breaking news as it’s released, market data/stock prices down to second/minute granularity, SEC filings with a daily-updated index, clinical trials as they post, economic data, and more all through one unified search API built for agents. If your AI systems are still relying on retrieval that lags by days or weeks, you’re not building real-time systems you’ve just built a guessing machine. Stop patching over the problem. Fix the pipeline. Integrate our API.
-
-
This is the DeepSeek moment of Search? A small team of cracked engineers built a search API for AI agents that outperforms everything built with 20x the funding. AI systems are about to become the biggest users of information on Earth. They don’t need a better Google. They need a retrieval layer built for reasoning, scale, and access beyond the web. That’s what we’re building.
-
Valyu reposted this
Join us for our AI Builders meetup next week that we’re cohosting with CRV alongside our friends at Valyu and Lemma (YC F25).
One of my favorite things about being in the world of startups and venture are the friendships over the years and in this case the last 5. I met Caitlin Bolnick Rellas from CRV back in 2020 through fellow Penn alum Jeremy Zhu. Caitlin was still at her previous firm and hadn’t even joined CRV yet. A few weeks after we met, we immediately collaborated on our first co-investment together. We’ve since been able to collaborate on several more things over the last 5 years since. Shortly after we closed our first fund in 2021, she gifted me this ‘Comma Capital’ mug (my first and only piece of Comma swag). And she has been one of our biggest supporters over the last few years through the ups and downs of building Comma. Next week, we’re cohosting an event with Caitlin and the CRV team alongside our friends at Valyu and Lemma (YC F25) to bring great AI engineers and builders in San Francisco to one place. We hope that folks that attend our event meet atleast one new friend. If you’re interested in joining us, comment ‘SF friendships’ and can send you the link!
-
-
Valyu reposted this
🚀 𝗟𝗮𝗻𝗰𝗲𝗗𝗕 𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 𝗡𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 𝗔𝘂𝗴𝘂𝘀𝘁 𝗨𝗽𝗱𝗮𝘁𝗲𝘀 🎥 From Hollywood to code reviews, LanceDB is powering the future of AI data. 1️⃣ Netflix is building a 𝗠𝗲𝗱𝗶𝗮 𝗗𝗮𝘁𝗮 𝗟𝗮𝗸𝗲 on LanceDB to unlock next-gen analytics and ML on media assets. 2️⃣ CodeRabbit is delivering sub-second AI-powered code reviews at scale, thanks to LanceDB’s multimodal performance. 3️⃣ With #Lance Namespace, teams can now manage Lance tables seamlessly across Apache Hive, Unity Catalog, Amazon Web Services (AWS) #Glue, and more. 💡 In case you missed it, August's new #engineering #blog: 📔 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 with Geneva 📕 Columnar File Reader in Depth – 𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗮𝗹 𝗘𝗻𝗰𝗼𝗱𝗶𝗻𝗴 📗 LanceDB WikiSearch: Native 𝗙𝘂𝗹𝗹-𝗧𝗲𝘅𝘁 𝗦𝗲𝗮𝗿𝗰𝗵 on 41M Wikipedia 🌍 From the community: #Lance now supports #GEO 𝘁𝘆𝗽𝗲 — thanks to incredible contributors from ByteDance @ddupg and Uber @jaystarshot for making it possible. In Aug, we also went for a small trip across #Berlin, #Amsterdam, and #London. Thanks for the support and collaboration from dltHub, PyData Berlin, Databricks Amsterdam, Valyu, Amazon Web Services (AWS) London, #VLDB. 👉 Read the latest newsletter for a full Aug roundup: https://lnkd.in/dMKVaBgM #AI #DataInfrastructure #MachineLearning #LanceDB
-