What is Exa?
Exa is an AI search API that gives AI agents and applications real-time access to web data for developers and engineering teams building LLM-powered products. It covers search across the full web, news, companies, people, research papers, and code. The API returns results in multiple formats: standard URL results, token-efficient highlights, LLM-generated summaries, and structured JSON outputs. Exa also includes a web crawling API for extracting content from any URL and a Websets product for building filtered lists of companies or people from the live web. Developers can access the API through Python and JavaScript SDKs or connect through an MCP server.
Exa Video
Features & Benefits
- Web Search: Query the live web and return real-time results across multiple categories including news, companies, people, research papers, and code as indexed content.
- Deep Search: Run complex multi-step research queries and get structured results with grounded citations. This AI search API mode supports configurable latency profiles from a few seconds up to roughly 20 seconds for reasoning-intensive tasks.
- Content Extraction: Fetch full page text, targeted highlights, or LLM-generated summaries from any URL, including JavaScript-rendered pages and PDFs.
- Highlights: Extract the most relevant passages from source pages using a specialized model trained for this task. Highlights reduce token usage by up to 90% compared to passing full page text to an LLM.
- Structured Outputs: Return search and content results as JSON matching a schema you define. Fields, types, and required properties are all configurable per request.
- AI Summary: Generate an LLM-produced answer or abstract directly from search results. Output can be streamed to the client.
- Websets: Build filtered lists of companies or people using natural language criteria. Results are verified by AI agents against multiple data sources before delivery. Export as CSV or connect via the Websets API.
- Subpage Crawling: Crawl linked subpages within a site in a single request. Target specific sections like documentation or about pages using the subpageTarget parameter.
- Monitors: Schedule recurring searches and receive results at a webhook endpoint. Each run deduplicates against prior runs so only new content gets delivered.
- Search Verticals: Access specialized indexes for company data, people data, news, code, and research. Each vertical is optimized for queries in that category.
- Latency Modes: Choose between Instant (under 180ms), Fast (around 450ms), Auto (roughly 1 second), and Deep (4 to 12 seconds) depending on the speed and depth the task requires.
- Content Freshness Control: Set cache behavior per request with the maxAgeHours parameter. Options range from always-live crawl to cache-only retrieval.
- MCP Server: Connect the Exa AI search API directly to AI assistants and coding agents that support the Model Context Protocol.
- Python and JavaScript SDKs: Integrate the API into existing projects using official libraries for both languages.
- Integrations: Connect with compatible AI frameworks and third-party tools through the Exa integrations catalog.
- Zero Data Retention: Opt into a configuration where all queries and response data are purged automatically per your requirements.
- SOC 2 Type II: Exa maintains SOC 2 Type II certification across its infrastructure.
What can Exa do?
- Search the web in real time from an AI agent
- Extract structured data from any website
- Crawl a full documentation site with one API call
- Get token-efficient highlights from long web pages
- Run deep research queries with cited structured outputs
- Build a filtered list of companies matching custom criteria
- Monitor a topic and receive new results via webhook
- Search for recent news about a company or industry
- Find people data and company information from the live web
- Fetch content from JavaScript-rendered pages
- Stream an AI-generated answer from web search results
Real-World Applications
Development teams building AI agents often need those agents to pull current information from the web without writing custom scrapers. Exa’s AI search API handles that retrieval layer directly. An agent built for a coding assistant can query the Exa API for the latest library documentation or GitHub activity and get back clean, structured content ready to pass to an LLM.
Sales and marketing teams building data products may need filtered lists of companies that match specific criteria, like funding stage, headcount growth, or industry signals. Websets lets a team describe those criteria in plain language and receive a verified list with custom enrichment columns. The output can go straight into a CRM or outbound sequencing workflow via CSV export or the API.
Research-oriented applications that track regulatory changes, funding announcements, or scientific publications can set up Monitors to run recurring searches on a schedule. Each run deduplicates against prior results and delivers only new content to a webhook endpoint. A compliance team tracking policy updates or a fund tracking investment activity in a sector could automate this kind of ongoing research without manual query work.
Products that need to process content from many URLs at once, like document review tools or competitive intelligence platforms, can use the AI search API’s content extraction and subpage crawling capabilities. A tool analyzing an entire knowledge base or competitor documentation site can pull structured text across many linked pages in a single request, with freshness controlled at the parameter level.
