Exa
✓ Editorially verifiedWeb search API built for AI agents, with structured outputs and token-efficient highlights.
Pick Exa if you are wiring web retrieval into an AI agent and want a single API that returns clean, schema-shaped, token-efficient results.
Skip it if you need a self-hosted or open-source search stack, or your retrieval is purely over internal documents.
Exa (formerly Metaphor) is a web search and retrieval API designed for AI agents and LLM applications that need fresh, structured web data. It offers fast keyword-and-neural search, automated crawling, JSON outputs against custom schemas, and a Highlights mode that extracts the relevant excerpts of a page so callers can cut up to 90% of tokens they would otherwise feed to a model. A Deep Research mode lets you trade latency for thoroughness.
The target customer is anyone building retrieval into an agent stack: Cognition, Cursor, HubSpot and Monday.com are cited on the homepage. Beyond general web search, Exa has vertical indexes for companies, people, and code repositories, which is the differentiator versus rolling your own SerpAPI + scraper pipeline. Pricing is usage-based with a free playground; production tiers are paid and enterprise plans include SSO, SOC 2 Type II, and zero-retention options.
It is proprietary and API-only, with an MCP server for plug-in use from Claude, Cursor and similar clients. If you want a pure open-source self-hosted search, look elsewhere; if you want managed retrieval that drops into an agent in an afternoon, this is one of the cleaner options.
Exa has quietly become the default web-search layer for serious agent builders, and the rebrand from Metaphor reflects that maturity. The Highlights and structured-output features pay for themselves in token savings alone, and the vertical indexes are a real moat. Worth a benchmark against your current SerpAPI + scraper stack.
— The AI Tool Bible editorial team
Pros
- ✅ Purpose-built for LLM/agent use, not retrofitted consumer search
- ✅ Highlights mode dramatically cuts tokens sent to the model
- ✅ Structured JSON outputs against custom schemas
- ✅ Vertical indexes for companies, people, and code
- ✅ SOC 2 Type II with zero-retention enterprise option
Cons
- ⚠️ Proprietary, closed source
- ⚠️ Pricing not transparent on homepage beyond the free playground
- ⚠️ Coverage and freshness depend on Exa's crawl, not yours
Use cases
Explore related
Compare with similar tools
All in RAG →Pinecone
FeaturedManaged vector database for production-scale similarity search.
LlamaIndex
FeaturedData framework for connecting LLMs to your data.
Weaviate
Open-source vector DB with hybrid search and modules.
LangChain
The broad LLM application framework — chains, agents, retrievers.
Vespa
Yahoo's open-source search engine with vector + sparse retrieval.
Chroma
Embedded, developer-friendly vector store for Python.