Exa

An AI-native search engine that uses neural networks to understand search intent, returning structured results instead of lists of links.

Exa (formerly Metaphor) is an AI-native search engine, upgraded to version 2.0 in 2026. Unlike the keyword matching of traditional search engines (Google/Bing), Exa uses neural networks to understand search intent and returns structured results (extracted entities, relationships, and summaries) rather than simple lists of links. For example, when you search for “the most noteworthy AI startups of 2026,” Exa directly returns a list of companies along with funding status, founders, and product overviews—rather than a pile of news links. Exa’s core technology is “neural retrieval”: web content is encoded as vectors, and queries are matched via semantic similarity rather than inverted indexes. This allows it to far outperform traditional search engines when handling complex queries (such as “find all open-source databases written in Rust”).

The free tier includes 1,000 queries per month; Pro is $50/month for 10,000 queries plus API access; Enterprise is custom-priced (unlimited queries + dedicated index + SLA). Exa’s pricing sits in the mid-range among search APIs, but its semantic understanding capabilities make it irreplaceable in specific scenarios.

In terms of ecosystem positioning, Exa competes with Perplexity, Google, and Brave Search. Perplexity targets consumers (conversational search), Google targets the mass market (keyword search), while Exa targets developers and enterprises (API-first + structured results). Its typical users are AI application developers (who need a high-quality search API), researchers (who need large-scale literature retrieval), and enterprises (who need internal knowledge base search). It is well suited for search scenarios that call for semantic understanding rather than keyword matching.

Note: The free tier has a limited number of queries; Chinese-language search quality lags behind English; real-time indexing carries a 1-2 hour delay; some complex queries may take multiple attempts to yield ideal results.

Source: Agentopia · Full-site machine catalog available at /llms.txt