Loading tool details…
Neural search API for LLMs and agents — semantic web search, live crawling, structured extraction, and a Websets research product.
Neural search API for LLMs and agents — semantic web search, live crawling, structured extraction, and a Websets research product.
Exa (formerly Metaphor) is a search infrastructure company built for LLMs and agents rather than humans. The core product is a search API that supports neural (embedding-based) queries, keyword queries, and hybrid, over a proprietary index of the public web that Exa continuously re-crawls. Results include content — full page text, cleaned HTML, or auto-extracted structured fields — so an agent can retrieve and consume the source in a single call rather than searching and then fetching. Exa's differentiators over commodity search APIs are (a) neural search that finds pages by 'what they mean' rather than keyword overlap, useful for exhaustive research prompts, (b) 'similar to this URL' search for finding lookalike companies, papers, or products, and (c) Websets — a research product that runs a slow, exhaustive multi-hour search across the internet to build labelled datasets on demand. Exa is used inside Perplexity, Anthropic Claude, and countless RAG stacks and agent frameworks.
Exa's neural search engine excels at finding content that keyword-based search APIs miss entirely, making it invaluable for AI agents doing research and discovery. The smaller index means some coverage gaps, but semantic search quality is often superior to traditional alternatives.
A purpose-built embeddings-based web index that ranks results by semantic meaning, allowing AI agents to issue natural-language queries and retrieve conceptually relevant pages even when keywords do not match.
Fetches clean, parsed text, summaries, and highlights from any URL returned by search (or supplied directly), removing the need for custom scraping, boilerplate stripping, and HTML cleanup in RAG pipelines.
Returns a synthesized natural-language answer with inline citations from the open web, enabling developers to embed Perplexity-style cited responses into their own applications via a single API call.
Higher-level product that turns a natural-language description of an entity type into a structured, spreadsheet-style dataset of matching results enriched with custom columns — useful for lead generation, recruiting, and market research.
Orchestrates multi-step search, browsing, and synthesis to produce long-form, cited research reports from a single high-level prompt, designed for agentic workflows that need depth rather than a single SERP.
Official Model Context Protocol server plus SDKs and integrations let MCP-aware clients (Claude, Cursor) and frameworks (LangChain, LlamaIndex) call Exa as a native tool without custom adapters.
Supports filtering by domain, date range, language, and content type, with an auto mode that dynamically chooses between neural and keyword search per query for optimal recall and precision.
Prompt builder, API dashboard with usage analytics, status page, and detailed documentation reduce time-to-first-query and make it easier to monitor production deployments.
$0
$10/month
Usage-based
Custom
Ready to get started with Exa?
View Pricing Options →Exa works with these platforms and services:
We believe in transparent reviews. Here's what Exa doesn't handle well:
Exa has continued to expand beyond a pure search API into a broader AI web-data platform. Recent additions include the Websets product for building structured datasets from natural-language descriptions, a Deep Research API for long-form cited reports, and an official MCP server that lets Claude, Cursor, and other Model Context Protocol clients call Exa as a native tool. The Answer endpoint has been hardened for production use with citation support, and the Contents API now offers richer summaries and highlights. Developer tooling has grown with a prompt builder, an updated API dashboard, and additional case studies and integration partners, reflecting Exa's positioning as core retrieval infrastructure for agentic applications.
Web Search & Research
One secure web-access API for AI agents — real-time search, extract, crawl, and research endpoints with 180ms p50 latency and built-in PII / prompt-injection guardrails.
AI Agent
Brave Search API provides enterprise-grade web search access to an independent index of over 35 billion pages — one of the few search APIs not built on Google or Bing results. This independence makes it valuable for AI applications that need unbiased, diverse search results for retrieval-augmented generation (RAG), agent tool use, and training data collection. The API recently eliminated its free tier, replacing it with $5/month in free credits (~1,000 queries) on paid plans. Search queries are
Search API
Fast, cheap Google Search API for AI agents — sub-2s responses at pennies per thousand queries.
Pricing and features change often. If anything on this Exa page is wrong or missing, tell us and include a link to the source. We review every correction.