This new type of search engine is shaking up our habits 🔍

What if the web became a database, readable like an open book? A tech startup, Exa, has set this ambitious goal with its search engine, Websets. Promising to transcend the logic of keywords, this innovation could reshape how we navigate online.


With nearly two billion websites competing for our attention, finding precise information can sometimes feel like an impossible quest. Traditional tools, like Google, rely on rankings based on keywords. While these systems are efficient for simple searches, they often fail to meet more complex demands. This is where Exa Websets steps in.

This search engine, still in its launch phase, uses a completely new approach: embeddings. Unlike traditional methods, this technology encodes information by grouping it based on meaning and context, rather than associating specific terms. The result? Answers that go straight to the point.

Will Bryk, CEO of Exa, explains that the method used by Websets excels at handling particularly complex queries. During a demonstration, the engine responded to a detailed inquiry about "all PhD students who have worked on developer-focused products, graduated from prestigious universities, and maintain a blog." Websets generated an exhaustive list of names, along with links to their LinkedIn profiles. This level of precision, unattainable with traditional search engines, illustrates its ability to combine refined semantic searches with advanced contextualization of results.

However, this precision comes at a cost. Exa had to encode billions of web pages—a titanic task, though still modest compared to Google's colossal indexes. Each search requires impressive computing power, sometimes extending the wait time to several minutes. This is a trade-off users are willing to accept in exchange for tailor-made results.


By incorporating techniques from advanced language models, Websets goes beyond simple word predictions to anticipate the most relevant connections. According to Andrew Gao, a computer science student, this tool is particularly useful when precise terms are lacking. However, for direct questions, like an isolated fact or definition, traditional search engines remain faster.

Some experts remain skeptical. The embeddings technology, while powerful, can lead to information loss. Bryk is aware of this but asserts that his team is actively working to overcome these limitations. The ultimate goal: to transform the search experience into a process as structured as it is efficient.

Though the path to an organized web remains long, Exa offers a glimpse of what the future might hold: a search engine that not only searches but comprehends. A slow but promising revolution.

What is the embeddings method used by Exa Websets?

The embeddings method is based on the principle of digitally encoding textual information to capture its contextual meaning. Each word, phrase, or document is transformed into a series of mathematical values that reflect their semantic relationships.

For example, closely related terms like "researcher" and "scientist" are represented by similar values, making their association easier. This representation allows the algorithm to understand not only the words themselves but also their meaning in a given context.

In practice, this means that the search engine can establish relevant connections between complex queries and web data, even in the absence of exact keywords.

Finally, this method is particularly useful in cases where thematic relationships and multi-level queries need to be explored, surpassing the limitations of simple keyword matching.