Exa API

Performs real-time web search using neural embeddings to understand query intent and semantic meaning rather than keyword matching. Returns ranked results with full page content (not snippets) and relevance highlights. Supports three latency profiles: Instant (<180ms), Auto (~1s), and Deep Search (up to 60s) for varying use cases. Integrates directly with AI agent frameworks via tool-calling APIs for Claude, GPT, and other LLMs.

Performs complex multi-step web research with structured output extraction and reasoning. Accepts complex queries and returns organized, citation-backed results with extracted structured data. Latency up to 60 seconds allows for iterative search refinement and content synthesis. Designed for research tasks requiring more than simple keyword matching, such as comparative analysis, fact-checking, or data aggregation across multiple sources.

Supports filtering search results by domain inclusion/exclusion lists and source restrictions. Allows developers to limit searches to specific domains (e.g., only news sites, only GitHub) or exclude domains (e.g., exclude social media). Filtering is applied server-side, reducing irrelevant results and improving result quality for domain-specific queries.

Extracts structured data from search results and web pages with citations linking each extracted field back to source URLs. Enables building applications that return organized, verified data instead of raw search results. Works in conjunction with Deep Search for complex extraction tasks. Supports custom schema definition for domain-specific data extraction.

Supports retrieving and processing content from multiple URLs or search results in batch operations. Enables efficient processing of large numbers of pages without individual API calls per page. Batch operations are optimized for throughput and cost efficiency, making them suitable for large-scale content processing pipelines.

Provides enterprise-grade features including Zero Data Retention (ZDR) option for privacy-sensitive applications and tailored content moderation policies. ZDR ensures no query or result data is retained by Exa after request completion. Custom moderation allows enterprises to define content policies specific to their use case. SOC 2 Type II certified for security and compliance.

Provides enterprise-grade security features including SSO (Single Sign-On) for authentication, Zero Data Retention (ZDR) for privacy-sensitive deployments, and SOC 2 Type II compliance certification. Enables enterprise customers to meet security and compliance requirements without custom integration or data handling agreements.

Provides interactive API dashboard at dashboard.exa.ai with guided onboarding that generates stack-specific integration code based on user's technology choices. Dashboard handles API key generation, SDK installation, and provides code examples for selected framework/language combination. Reduces setup time from hours to minutes.

Retrieves complete HTML/text content from web pages referenced in search results or provided URLs. Supports selective highlighting of relevant passages to reduce token usage in LLM context windows. Highlights are computed based on query relevance, allowing LLMs to focus on pertinent sections without processing entire page text. Configurable to return different content types (full text, HTML, markdown) and supports batch retrieval of multiple pages.

Monitors the web for new content matching specified queries at scheduled intervals (daily, weekly). Delivers new results via webhooks to a specified endpoint when matches are found. Enables continuous tracking of web events, news, competitor activity, or other time-sensitive information without polling. Results are delivered asynchronously with full page content available for each match.

Generates direct answers to queries by searching the web and synthesizing information from multiple sources in real-time. Supports streaming responses for progressive answer delivery. Answers include citations linking back to source URLs, enabling verification and transparency. Designed for use cases where users need quick, sourced answers rather than raw search results.

Provides specialized search indexes optimized for specific content types: People (person search), Companies (70M+ structured company database with fields like company_name, ceo_name, founded_year), and Code (GitHub repos, Stack Overflow, documentation). Each vertical maintains structured metadata enabling filtered search and extraction of specific fields without full-page content retrieval.

Automatically generates AI-powered summaries of web pages to reduce token usage in LLM context windows. Summaries are computed server-side and returned alongside full content, allowing applications to choose between full text and condensed summaries based on use case. Pricing at $1 per 1k pages makes it cost-effective for large-scale content processing.

Provides native integrations with major AI frameworks and LLM providers via tool-calling APIs. Supports Anthropic Claude tool calling, OpenAI function calling, Vercel AI SDK, LangChain, CrewAI, and LlamaIndex. Integrations handle schema generation, parameter marshaling, and response parsing automatically, eliminating boilerplate code for agents.

Provides an MCP (Model Context Protocol) server implementation enabling Claude and other MCP-compatible clients to access Exa search capabilities. Allows Claude to use Exa as a native tool without explicit function calling setup. Supports both Exa MCP and Websets MCP for different use cases.

Offers three configurable latency profiles for different use cases: Instant (<180ms for real-time agent loops), Auto (~1s for balanced performance), and Deep Search (up to 60s for complex research). Allows developers to trade off latency for result quality and reasoning depth. Instant mode is optimized for agent tool calls with minimal latency overhead.