Exa

Exa is a search API built specifically for AI agents and RAG pipelines, not for human keyword queries. It uses neural ranking to find semantically relevant content, returns clean machine-readable text with sub-200ms latency, and supports specialised verticals like code search, company research, and real-time monitoring. Exa AI raised $85M in Series B funding in 2025 and serves developers building everything from research agents to data enrichment pipelines.

What Exa does:

• •Neural search a meaning-based ranker finds semantically relevant pages instead of brittle keyword matching, which is critical when an LLM agent does not know the exact wording of what it needs.
• •Sub-200ms Exa Instant a low-latency tier launched in 2026 returns search results in under 200ms, eliminating the bottleneck that wrecks real-time agent workflows on slower APIs.
• •Structured machine-readable output responses ship as clean text, highlights, and metadata you can drop straight into an LLM prompt or a vector store. No HTML scrubbing, no boilerplate cleanup.
• •Search-with-contents in one call the API returns full page content alongside the ranked list, so you do not have to make a second crawler request to read each result.
• •Specialised verticals dedicated modes for code search (across GitHub and docs), company research, research papers, and real-time web monitoring.
• •Deep Search for multi-step research a higher-cost tier handles multi-hop questions where the agent reads, reasons, and re-queries in a single API call.
• •MCP server support expose Exa as a Model Context Protocol endpoint so Claude, Cursor, and other agent frameworks can call neural search natively.
• •Browser extension and dashboard a Chrome extension and a web playground let developers test queries, capture page data, and tune retrieval before wiring into production code.
• •OpenAI-compatible API and SDKs Python and TypeScript SDKs plus a documented REST endpoint, so existing client tooling drops in without rewrites.

Pricing:

• •Free $0/mo: 1,000 requests per month with $10 of free credits to test paid tiers, no card required to sign up.
• •Pay-as-you-go API (billed monthly): Exa Instant at $5/1,000 requests, Search-with-contents at $7/1,000 requests (10 results per call), Deep Search at $12/1,000 requests, content retrieval at $1/1,000 pages, billed /mo based on usage.
• •Websets / Pro from $49/month: 8,000 credits per month for high-volume agent pipelines, plus higher rate limits and priority support for production retrieval workloads.

Limitations:

• •Developer-only product Exa is a raw API, not a consumer search box. To use it, you write code or hire someone who can. Non-technical operators cannot point and click their way through it.
• •Costs scale fast at volume the 1,000-request free tier evaporates quickly. Search-with-contents at $7/1k and Deep Search at $12/1k add up if you run daily research pipelines without caching aggressively.
• •Weak on social and forum content Exa's neural index excels at structured content (blogs, docs, research papers) but underperforms on Reddit, X, and other community-sentiment surfaces. Pair with another source if those matter.
• •Inconsistent data freshness crawl recency varies by domain. If your application depends on news, real-time pricing changes, or last-week launches, validate timestamps in production rather than trusting the index.
• •Rate limits cap throughput the free tier and lower paid plans cap at roughly 10 queries per second. High-volume automation needs the Websets tier or enterprise contact-sales pricing to clear.