Embeddings based web search

Exa ranks web pages by meaning rather than exact keyword match, using embeddings to judge how closely a page fits the intent behind a query. A Beam agent submits a request when a workflow needs outside context, such as researching a company mentioned in a deal or checking background on a topic before a call. The agent reads the matched pages back, drops the relevant ones into the record it is updating, and leaves anything ambiguous, promotional, or contradictory for a person to confirm before it appears in a customer facing document or gets used in a decision.

Full text page retrieval

Exa returns the actual content of the pages it finds, not just titles and links, so an agent is not stuck guessing what a result contains. It can pull that text directly into a summary, a brief, or an enrichment field instead of asking someone to open each page and copy details by hand. This matters most in research heavy steps, where an agent gathers background before a meeting or drafts a first pass at a company profile. Anything that reads as outdated, promotional, or inconsistent with other sources gets set aside for a human reviewer rather than saved automatically into the record.

Result relevance ranking

Exa scores every result by how closely it matches the meaning of the query, so the top entries tend to be the ones worth reading rather than the ones with the most overlapping words. A Beam agent uses that ranking to limit what it reviews, working through the first few results instead of a long list of links. Where the top results disagree with each other or the subject is sensitive, such as legal or financial claims about a company, the agent leaves the page unread and passes the query to a person to judge before anything is recorded.