Exa
by Exa Labs
The search engine for AI
Exa is a search API that gives AI agents live web data from its own 100-billion-document index rather than a reseller's. Enterprise developers use it to ground retrieval, run multi-step research and build enrichment pipelines, replacing brittle scraping and generic SERP wrappers with semantic search, token-efficient content extraction and cited answers.
Exa is a web search API built for AI systems rather than human browsers, operating what the company describes as an independent index of roughly 100 billion documents and 1.4 trillion tracked URLs. It was founded in 2021 by Will Bryk and Jeff Wang as Metaphor Systems, came out of Y Combinator's S21 batch, and rebranded to Exa in January 2024. It now sells a family of REST endpoints that agent developers compose directly: Search, for neural and keyword retrieval across web, news, company, research, people and financial verticals at configurable latency from roughly 180 milliseconds to ten seconds; Contents, for page extraction, whose AI-selected highlights mode the company says cuts token usage by about 90 percent; Answer, for cited LLM responses; Monitors, for recurring scheduled queries; Websets, which returns large structured result collections enriched into table columns across more than 70 million companies; and an Agent API that runs multi-step research at five fixed-price effort levels or in metered auto mode. Official Python (exa-py) and JavaScript (exa-js) SDKs, an MCP server with roughly 4,900 GitHub stars, and a ChatGPT/Codex plugin are the main integration paths. Exa reports more than 500,000 developers and names Cursor, Cognition, HubSpot, Mozilla's Firefox and AWS among its users, with query volume growing from about 100 million a month in April 2025 to roughly 1 billion in April 2026. Sacra put revenue at about $10 million as of September 2025. The company raised $85 million at a $700 million valuation in September 2025 led by Benchmark, then $250 million led by Andreessen Horowitz at a $2.2 billion valuation in May 2026. It sits between AI-native search APIs such as Tavily, Brave and Jina AI and the built-in web tools of OpenAI, Anthropic and Google.
Engineering leaders whose agents or RAG pipelines need current, citable web context and who are paying too much in tokens to feed raw scraped pages into a model.
One API covering search, extraction, cited answers and multi-step research over an index Exa owns, with highlights that cut roughly 90 percent of the tokens a retrieved page would otherwise cost.
At a Glance
- Category
- Enterprise Search & Knowledge
- Pricing
- Usage-based, Freemium, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, AI Engineers, Data Scientists, Product Engineering Leads
- Deployment
- API-based, Cloud-only
- Founded
- 2021
- Headquarters
- San Francisco, United States
- Customers
- 500,000+ developers (vendor-stated); named users include Cursor, Cognition, HubSpot, Mozilla Firefox and AWS
Key Features
- ✓Independent web index
Exa crawls and embeds its own index of roughly 100 billion documents and 1.4 trillion tracked URLs, so results do not depend on reselling Google or Bing.
- ✓Neural and keyword Search API
Semantic embedding search across web, news, companies, research papers, people and financial data, with latency configurable from about 180ms to ten seconds.
- ✓Contents API with highlights
Extracts page text and returns AI-selected highlights that the vendor says remove roughly 90 percent of tokens before content reaches your model.
- ✓Agent API with fixed effort tiers
Runs multi-step research autonomously at five fixed-price effort levels from $0.012 to $1.00 per request, so cost per task is knowable in advance.
- ✓Websets structured collections
Builds verified, table-shaped result sets with enrichment columns across more than 70 million companies, replacing manual list building and prospecting research.
- ✓Monitors for recurring queries
Reruns a saved search on a schedule and surfaces only newly published matches, for competitive tracking and continuous signal collection.
- ✓MCP server and official SDKs
Python (exa-py) and JavaScript (exa-js) SDKs plus an MCP server let Claude, ChatGPT and Codex call Exa as a native tool.
Capabilities
Use Cases
- •Grounding coding agents
Cursor and Cognition's Devin call Exa so coding agents read current library documentation instead of relying on stale training data.
- •Automated deep research
Analysts trigger the Agent API to run multi-step web research and return cited findings without a human driving individual searches.
- •Token-efficient RAG over the live web
Contents highlights shrink retrieved pages by roughly 90 percent of tokens, cutting inference spend across long-context grounding pipelines.
- •Go-to-market list building and enrichment
Websets returns verified company and people lists as tables with enrichment columns, replacing manual prospecting research inside revenue teams.
- •Competitive and regulatory monitoring
Monitors reruns saved queries on a schedule and surfaces only newly published matches for compliance watch and competitor tracking.
Ideal For
Best For
- ✓Teams building coding or research agents that need current web context instead of model training data
- ✓RAG pipelines where the token cost of retrieved pages is the dominant inference expense
- ✓Semantic and entity-shaped queries ('startups doing X in market Y') that keyword SERP APIs handle badly
- ✓Product teams embedding company or people enrichment directly into their own application
- ✓Developers who want one vendor for search, extraction, answers and multi-step research rather than four
Not Ideal For
- ✗Buyers needing exhaustive breadth — Exa's index is materially smaller than Google's, so long-tail coverage of obscure pages can lag general-purpose SERP APIs
- ✗Regulated buyers who need HIPAA or zero data retention without an Enterprise contract; both are Enterprise-only, and GDPR, ISO 27001, SSO and data residency are not documented on the security page
- ✗Organisations with strict content-provenance policies — Exa has announced no publisher licensing agreements while selling access to crawled content, and has been publicly accused of indexing sites that disallow it
- ✗Teams that need a fixed, predictable monthly bill: six endpoints meter separately and agent runs bill in compute units
Integrations
Deployment
Market & Ratings
500,000+ developers (vendor-stated); named users include Cursor, Cognition, HubSpot, Mozilla Firefox and AWS
Market Analysis
Pros
- ✓Owns its index, so pricing, ranking and availability are not dependent on a Google or Bing reseller relationship
- ✓Highlights and structured extraction cut token spend downstream, which is usually where the real cost of web grounding sits
- ✓Strong production references — Cursor, Cognition, HubSpot, Mozilla Firefox and AWS — behind roughly 1 billion queries a month as of April 2026
- ✓$20 signup credit plus $10 in free credits every month makes a genuine technical evaluation free
- ✓SOC 2 Type II certified with a public trust centre offering SOC 2 reports and data processing agreements
Cons
- ✗Six separate meters — search, deep search, contents, answer, monitors and agent compute units — plus per-email and per-phone enrichment surcharges make spend forecasting genuinely hard for a high-volume agent
- ✗Crawling ethics are contested: a January 2026 Hacker News submission alleged Exa indexes personal site data in disregard of robots.txt, and Media Copilot notes Exa has announced no publisher licensing agreements while selling access to crawled publisher content
- ✗Compliance is thin outside an Enterprise contract — SOC 2 Type II is the only published certification, HIPAA and zero data retention are Enterprise-only, and GDPR, ISO 27001, SSO and data residency are absent from the security documentation
- ✗The public review base is tiny and self-selected: 16 Product Hunt reviews all at 5.0, mostly from founders who ship Exa inside their own products, and no substantive G2 or TrustRadius profile exists
- ✗Sacra put revenue at about $10M as of September 2025 against a $2.2B valuation, so buyers are underwriting a very early company for what becomes critical retrieval infrastructure
Pricing
Free
$0
- ✓$20 in credits on signup (about 2,800 searches)
- ✓$10 in credits every month
- ✓Full API access
- ✓No card required to start
Pay-as-you-go
From $7 per 1,000 searches
- ✓Search $7/1k requests (10 results), $1/1k extra results
- ✓Deep search $12-$15/1k requests
- ✓Contents $1/1k pages per content type
- ✓Answer $5/1k requests
- ✓Monitors $15/1k requests
- ✓Agent $0.012-$1.00 per request or $0.10 per agent compute unit
Enterprise
Contact for pricing
- ✓Volume discounts
- ✓Higher rate limits
- ✓SLAs
- ✓Zero Data Retention
- ✓HIPAA support
List pricing is fully published and purely usage-based: $7 per 1,000 searches returning ten results, $1 per 1,000 additional results, $12-$15 per 1,000 deep searches depending on reasoning depth, $1 per 1,000 content pages or AI summaries, $5 per 1,000 Answer calls and $15 per 1,000 Monitors calls. The Agent API bills either at fixed effort tiers ($0.012 minimal to $1.00 xhigh per request) or metered at $0.10 per agent compute unit plus $0.005 per search, $0.02 per email and $0.07 per phone enrichment. New accounts get $20 in credits and free-tier accounts $10 monthly. Volume discounts, higher rate limits, SLAs and zero data retention are Enterprise-only and unpriced, so a high-volume agent workload has six independent meters to forecast.
Security & Compliance
Connect
Sources
This page was written from 9 sources, 4 on domains other than exa.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Contextual AI
Enterprise RAG and Agent Composer platform for grounded AI agents on technical documentation
Quill
Production AI agents on the SQL database you already run — no migration, no assembled stack
Twin1 AI
A permission-aware AI twin for every professional, networked so expertise moves across the firm
DeepJudge
Institutional-knowledge search and AI workflows for law firms, with no data migration