Hey HN! We're Will and Jeff from Exa, and we're excited to introduce Exa Websets, our embeddings-powered search engine that aims to deliver exactly what you’re looking for. You can now conduct precise searches for complex queries, such as "all startups working on open-source developer tools based in San Francisco, founded between 2021 and 2025." We started building Exa because we were frustrated by the declining quality of search results on platforms like Google. What once felt like a magical information portal now often steers users toward SEO-optimized clickbait. With Websets, we're taking a different approach. For each search, we perform multiple embedding searches across Exa’s extensive vector database of the web to identify relevant candidates. Then, we use agent-driven workflows to verify that each result matches your exact criteria. The precision of Websets stems from two key innovations. First, we train custom embedding models for our main search algorithm, moving away from traditional keyword matching methods. These models are specifically designed to return the type of entity you're looking for. For instance, if you search for “startups working in nanotech,” conventional search engines might produce listicles featuring nanotech startups. In contrast, Exa’s embedding models will bring up the actual startup homepages that align with the meaning of your query. Secondly, we leverage large language models (LLMs) to provide the necessary intelligence for verifying every result. Each piece of data and result is backed by supporting references to ensure validity. This process can take anywhere from a few minutes to several hours, depending on the complexity of your query and the number of results requested. However, we believe this investment of time is worthwhile for highly valuable searches. One of the unique features of Websets is its tabular format, which transforms the way we interact with web search results. Rather than presenting them in a list, Websets organize results in tables. Users can enrich these tables by adding columns to find more specific information, like “number of employees” or “does the author have a blog?” The cells load asynchronously, making the process efficient and user-friendly. We hope this format makes the web feel more like a structured database. Here are a few example searches to showcase the capabilities of Websets: “Math blogs created by teachers from outside the US”: Link "Research paper about ways to avoid the O(n^2) attention problem in transformers, where one of the first author's first name starts with 'A', 'B', 'S', or 'T', and it was written between 2018 and 2022": Link “US-based healthcare companies with over 100 employees and a technical founder”: Link “All software engineers in the Bay Area with startup experience, proficiency in Rust, and a history of publishing technical content”: Video Demo You can try out Exa Websets at websets.exa.ai and explore the API documentation at docs.exa.ai/websets. We would love to hear your feedback and suggestions as we continue to refine and improve Websets!