Exa

Exa is a web search API built for AI agents, and this entry covers its Python SDK, which wraps the Search, Contents, Answer, and Agent APIs in synchronous and async clients. Search runs over Exa's index of the public web, with categories such as news, companies, research papers, people, and financials, and returns either result lists, a generated answer, or structured output.

Searches accept a system prompt and an output schema on every search type, returning plain text or structured JSON (nested up to two levels, at most ten properties). Search types range from a fast mode to deep variants (deep-lite, deep, deep-reasoning) that also take additional queries, and streaming search yields OpenAI-style chat completion chunks. Contents fetches page text or highlights, which are short excerpts that a dedicated model selects as the most relevant parts of a page, to keep the tokens passed to a model low.

The SDK packages Exa as ready-made web search and get-contents tools for OpenAI (Chat Completions and Responses) and Anthropic tool-calling loops, with handlers that answer every tool call. The Agent API runs a research task against an output schema with an effort level and an optional cost budget, and Agent Monitors (in beta) keep a table of entities and fields fresh on a cadence, tracking dynamic fields from news on each refresh and supporting one-shot backtests over a past news window.