Master Metaphor with our step-by-step tutorial, detailed feature walkthrough, and expert tips.
Explore the key features that make Metaphor powerful for ai search workflows.
Exa's Search API returns results in under 180ms and supports structured output extraction, allowing developers to define JSON schemas and receive clean, typed data directly from search results. This enables complex enrichment workflows — such as extracting company names, CEO names, and founding years — from Exa's database of over 70 million companies without additional scraping or parsing steps.
The Highlights feature extracts the most query-relevant excerpts from web pages, reducing the amount of text that needs to be sent to language models by over 50%. This is critical for AI agent and RAG pipeline builders who need to balance context quality with token costs, ensuring LLMs receive focused, relevant information rather than entire web pages.
Exa maintains industry-leading, curated web indexes for specific use cases including people, companies, code documentation, financial data, and news. These specialized indexes provide higher quality and more relevant results than general-purpose search for domain-specific queries, making them ideal for vertical AI applications and targeted research workflows.
Websets allows users to programmatically build structured, filtered collections of entities from web data based on defined criteria. This product turns Exa's search capabilities into a dataset assembly tool, useful for lead generation, competitive intelligence, market mapping, and creating curated training data at scale without manual research.
Exa provides SOC 2 Type II certified infrastructure with customizable Zero Data Retention policies that ensure queries and results are automatically purged according to organizational requirements. Combined with Single Sign-On team management, these features make Exa suitable for compliance-sensitive deployments in healthcare, finance, and government-adjacent industries.
Yes, Metaphor rebranded to Exa (Exa Labs Inc.). The core technology remains the same — an AI-native search engine built on neural networks for semantic understanding. The rebrand reflects the company's expanded focus beyond search into structured data extraction, web crawling, and a full suite of API products including Search, Contents, Answer, and Websets. Existing Metaphor API users were migrated to the Exa platform.
Exa's search API is built specifically for AI agents and programmatic use cases, delivering results in under 180 milliseconds with built-in features like structured output extraction, semantic highlights, and dedicated vertical indexes for companies, people, and code documentation. Unlike Google's Custom Search API, which mirrors traditional keyword-based web search, Exa understands meaning and context, enabling queries based on conceptual similarity rather than exact keyword matches. Exa also provides token-efficient content extraction optimized for feeding results into large language models.
Websets is one of Exa's core products that enables users to build curated, structured datasets from web search results. Rather than returning a simple list of links, Websets allows you to define criteria and automatically assemble collections of entities (such as companies, people, or resources) that match your specifications. This is particularly useful for lead generation, market research, competitive analysis, and building training datasets where you need structured, filtered data at scale.
Yes, Exa offers enterprise-grade security and compliance features including SOC 2 Type II certification, customizable Zero Data Retention (ZDR) policies that automatically purge queries and data based on your requirements, and Single Sign-On for team authentication and authorization management. Companies like Databricks, OpenRouter, and Flatfile use Exa in production. Enterprise customers can contact the sales team for custom pricing, dedicated support, and tailored deployment configurations.
Exa's Highlights feature intelligently extracts the most relevant excerpts from web pages based on your specific query, rather than returning entire page contents. This can reduce token budgets and LLM costs by over 50% because you only send the most pertinent text to your language model instead of full documents. This is especially valuable when building AI agents or RAG (Retrieval-Augmented Generation) pipelines where every token sent to an LLM has a direct cost impact and where irrelevant context can degrade response quality.
Now that you know how to use Metaphor, it's time to put this knowledge into practice.
Sign up and follow the tutorial steps
Check pros, cons, and user feedback
See how it stacks against alternatives
Follow our tutorial and master this powerful ai search tool in minutes.
Tutorial updated March 2026