Exa

Exa at a glance

Exa is an AI-native semantic search and retrieval API, not a consumer search website. It gives developers web search, page-content retrieval, answers, similarity search, deep-research workflows, and enrichment tools that can be embedded in RAG systems and agents. Its strongest fit is an application that needs more than links: relevant passages, structured outputs, citations, filters, and access to fresh web content.

The core API includes /search, /contents, /answer, and /findSimilar. Search supports neural or keyword-plus-neural hybrid retrieval, with domain and date filters. The Contents endpoint can return full page content or highlights, helping teams provide LLM context without always sending entire pages. Exa also supports live crawling and configurable latency, including sub-200ms Instant options. For more involved work, its Agent API handles asynchronous deep research, list building, and enrichment; Websets and contact enrichment extend that capability for sourcing workflows.

Pricing and cost control

Pricing is usage-based rather than seat-based. The free tier starts with $20 in sign-up credits and provides $10 in ongoing monthly credits. Base Search costs $7 per 1,000 requests and includes up to 10 results with token-efficient page contents. Results above 10 add $1 per 1,000 requests, while AI page summaries cost $1 per 1,000 pages. The Contents endpoint is $1 per 1,000 pages for each content type. Answer requests cost $5 per 1,000, and Monitors cost $15 per 1,000 requests.

Deep Search is $12 per 1,000 requests, while Deep-Reasoning Search is $15 per 1,000; both retain the extra-result and AI-summary charges. Agent runs range from $0.012 for Minimal effort to $1.00 for X-high effort, plus $0.10 per Agent Compute Unit and $0.005 per agent search tool call. Email enrichment is $0.02 per contact and phone enrichment is $0.07.

That model is flexible for prototypes, but budget carefully in production. Spend can rise when a workflow combines search requests with Contents page retrieval, summaries, live crawling, agent tool calls, and enrichment. At scale, calculate the full request path rather than comparing only the headline Search price.

Where it works well

Exa is particularly practical for coding agents searching repositories, documentation, and Stack Overflow; research assistants that need cited results; chatbots grounded in current web material; and monitoring or company/people-enrichment systems. Structured outputs with citations are useful when downstream software must consume results reliably. Configurable freshness and latency let developers choose a faster Instant path or deeper retrieval and reasoning when the task warrants it.

Limits to weigh

Exa is not the right choice for someone wanting a polished consumer search UI. It is a developer product, so implementation and tuning are part of the deal. Its index coverage is smaller than Google's, and result quality can vary by query. Deeper live-crawl and reasoning modes trade speed for depth, while rate limits and capacity should be validated before committing a high-volume production workload. Teams whose priority is the broadest possible index or the lowest cost for commodity search at very high volume should compare alternatives.

Verdict

Exa is a capable retrieval layer for AI applications that need semantic web search, usable page context, citations, and agent-oriented workflows. It earns its value when those capabilities improve application outcomes; it is less compelling when simple, extremely high-volume link search is the only requirement.

Sources

• https://exa.ai/pricing
• https://exa.ai/docs
• https://zapier.com/blog/best-ai-search-engine/