Last updated 07-17-2026
Category:
Reviews:
Join thousands of AI enthusiasts in the World of AI!
Exa
Exa is a search API built for AI agents, not traditional keyword search. The company indexes hundreds of billions of web pages and exposes search, page contents, deep research, and monitoring through a single API so developers can give agents live web context without building their own crawl stack.
The product spans several endpoints. Search returns ranked URLs with optional highlights that strip pages down to the tokens an LLM actually needs. Contents fetches full page text. Deep Search runs multi-step research with structured outputs and citations. Agent handles async list building and enrichment. Monitors watches the web on a schedule and pushes updates through webhooks.
Latency is configurable by search type, from sub-200ms instant mode to deep modes that take several seconds for harder queries. Category indexes cover companies, people, research papers, news, and more. Customers include Cursor, Cognition, HubSpot, OpenRouter, and Monday.com, with over 400,000 developers on the platform.
One API covers search, page contents, deep research, agents, and scheduled web monitors
Search types from ~250ms instant to multi-step deep research with structured JSON outputs
Highlights extract only the relevant passages from a page, cutting token use by up to 90%
Category indexes span 50M+ companies, 1B+ people profiles, and 100M+ research papers
Python and JavaScript SDKs plus an MCP server for plugging search into agent frameworks
SOC 2 Type II certified with zero data retention and single sign-on for enterprise teams
Purpose-built for AI agents with highlights and structured outputs that reduce LLM token costs.
Wide latency range from sub-200ms search to deep multi-step research in one API.
Trusted in production by Cursor, HubSpot, Cognition, and other major AI products.
Category-specific indexes for companies, people, papers, and news go beyond generic web search.
Usage-based pricing across multiple endpoints can be hard to forecast without monitoring dashboards.
Deep search modes take several seconds, so they are not suited to real-time chat latency requirements.
No self-hosted deployment option; everything runs through Exa's hosted API.
Is Exa free to use?
Yes. Exa offers a free tier with up to 20,000 requests per month. Paid usage starts at $7 per 1,000 search requests, with separate per-endpoint pricing for Contents, Agent, Deep Search, and Monitors.
What programming languages does Exa support?
Exa provides official Python and JavaScript SDKs (exa-py and exa-js). You can also call the REST API directly with cURL or any HTTP client.
How is Exa different from Google or Bing search?
Exa is built for AI applications rather than human result pages. It returns clean page contents, token-efficient highlights, and structured outputs designed to feed directly into LLM context windows and agent workflows.
What search latency options does Exa offer?
Exa supports multiple search types tuned for different speed-quality tradeoffs: instant (~250ms), fast (~450ms), auto (~1s), deep-lite (about 4s), deep (4-15s), and deep-reasoning (12-40s).
Does Exa integrate with agent frameworks?
Yes. Exa ships integrations for LangChain, CrewAI, LlamaIndex, Vercel AI SDK, OpenAI tool calling, Anthropic tool calling, and many other frameworks. An MCP server is also available.