Exa (formerly Metaphor)

A 'neural search' API built for semantic, meaning-based retrieval rather than keyword matching.

What it does

Exa (formerly Metaphor) sits in the category of search APIs: a hosted service that takes a query and returns structured, machine-readable search results instead of leaving you to render and scrape a results page yourself. For an application or an AI agent, that structure — clean JSON fields instead of raw HTML — is the entire value proposition.

Strengths

• Embedding-based retrieval that surfaces conceptually related pages even without exact keyword overlap
• A dedicated 'find similar' endpoint for expanding from a known good source
• Built-in content retrieval so agents can get full page text alongside the URL

Best-fit use cases

Research agents and citation-heavy workflows where conceptual relevance matters more than exact keyword rank.

Pricing shape

Usage-based pricing per search and per content-retrieval call, with startup-friendly free credits to start. As with any usage-billed API, the right plan depends on your query volume and how often your application — or an autonomous agent calling it in a loop — actually needs a fresh search rather than a cached one.

Where it fits in an agent stack

When you're wiring a "web_search" tool into an LLM agent's function-calling schema, the provider you choose determines both the shape of the JSON your parsing code needs to handle and the latency budget for each turn of the agent loop. Exa (formerly Metaphor) is one of several reasonable defaults — see our comparison framework for how to weigh it against alternatives.