Exa.ai review: how we actually run it

About

What Exa.ai actually is.

Exa is a search API built for AI agents and scripts to query, not people clicking through a results page. Send it a request and it returns full page content, structured JSON, or a synthesized answer, depending on which endpoint you call. There's no dashboard. Every result comes back through code, so someone on the team has to write the integration before it does anything.

Billing is pay-per-request rather than a flat subscription, and price varies by endpoint. Exa also publishes its own accuracy numbers, including a score on a benchmark called FRAMES; that's Exa grading itself, not a third party, so weigh the comparison to competitors accordingly. Summaries and structured answers are AI-generated, which means they still need a check before anyone treats them as fact.

What it does

Search: returns ranked web results as data, not a list of links.
Contents: pulls clean page text and highlights instead of raw HTML.
Deep Search: runs multi-step lookups for harder research questions.
Monitors: watches queries or pages and calls a webhook on change.
Agent: handles multi-step tasks and can find emails or phone numbers.
Websets: curated search collections for narrower verticals like company data.
Enterprise plans add SOC 2 Type II compliance and zero data retention.

It's built to be called, not clicked.

In the system

Where it sits in the pipeline.

Runs well with

The tools we'd run next to Exa.ai, and what each pairing covers.

Claude CodeAgents built with Claude Code can call Exa's search API directly to give the agent live, structured web results instead of stale training data.n8nExa's API can be wired into an n8n workflow as a search/enrichment step feeding downstream nodes like CRM writes or email drafting.FirecrawlBoth are API-first web data tools; teams often use Exa to find and rank relevant pages, then Firecrawl to scrape structured content off the pages Exa surfaces.

ClayClay workflows can call out to Exa's search/answer endpoints as an enrichment source for account or contact research inside a table-based workflow.

What it costs

Pricing, without the fiction.

From the Exa.ai pricing page · verified July 17, 2026

Free

per month, usage-capped

Up to 20,000 requests per month for free
Web search tool calls for agents
Webpage text and highlights
Configurable latency: 180ms to 1s

Search

$7 per 1,000 requests

pay-as-you-go, per-request

Web search tool calls for agents
Webpage text and highlights
Configurable latency: 180ms to 1s
Base price includes up to 10 results

Contents

$1 per 1,000 pages per content type

pay-as-you-go, per-request

Full page contents for LLM context
Token efficient highlights
Configurable livecrawl policies
AI page summaries available at $1 per 1,000 pages

Agent

$0.012 to $1.00 per run

pay-as-you-go, varies by effort mode

Minimal effort: $0.012/request
Low effort: $0.025/request
Medium effort: $0.10/request
High effort: $0.50/request
X-high effort: $1.00/request

Enterprise

Custom pricing

custom / contact sales

Up to 1,000 results per search
Custom rate limits (QPS)
Enterprise-grade support
SLAs and MSAs, 1:1 onboarding
Zero Data Retention
Volume discounts available

The bill is driven by request volume across separate metered endpoints, not a flat subscription: Search runs $7 per 1,000 requests (10 results), Contents (fetching page text) is $1 per 1,000 pages per content type, and Deep Search or Monitors cost more per 1,000 calls ($12-15 and $15 respectively). Costs stack when a single workflow chains a search call with a contents call and an AI summary, since each is billed separately at its own per-1,000 rate. What looks cheap at prototype scale (the 20,000 free requests/month) can get harder to predict at production scale because usage is per API call, not per seat or per contact record, so a chatty agent workflow that calls Exa multiple times per user query multiplies costs fast. The Agent product's compute-unit pricing ($0.10/ACU, up to $1.00/request at X-high effort) adds a second, less predictable cost axis on top of the base search pricing.

The verdict

Choose it, or skip it.

Choose it if

Teams building AI agents or RAG pipelines that need live web search results as structured JSON rather than a link list, since Exa returns page content and can output structured answers.
GTM/data teams doing custom scraping or company research workflows where a pay-as-you-go API (no seats, no contracts) fits an engineering-driven build better than a per-seat scraping tool.
Anyone testing feasibility first: the free tier covers up to 20,000 requests a month, enough to prototype an agent workflow before committing spend.
Technical builders comfortable working directly with an API and reading usage-based billing, not marketers who want a point-and-click UI.

Skip it if

Non-technical teams that want a dashboard or no-code interface: Exa is API-only (interface listed as api), so someone has to write and maintain integration code.
Teams with predictable, high-volume steady-state scraping needs where a flat-rate tool would be cheaper than metered per-request pricing once volume is high.
Buyers who need guaranteed fixed monthly costs for budgeting; Exa's pay-as-you-go model means the bill moves with usage and can be hard to forecast without rate limiting.

Alternatives

If not Exa.ai, then what.

FirecrawlFree tier (1,000 credits/mo); paid plans from $16/mo (Hobby), up to $599/mo (Scale); enterprise custom quote.
Choose Firecrawl when the job is turning a known URL or site into clean structured content or markdown for an LLM, rather than discovering relevant pages across the open web first.
ApifyFree tier ($5/mo usage credit) then paid plans from $29/mo (Starter), usage-based overage billing
Choose Apify when you need a large marketplace of pre-built scrapers (social platforms, marketplaces, maps) with a visual actor store instead of writing search queries against a single API.
BlitzAPI
Choose BlitzAPI when the target is a specific data extraction task already documented for LLP's stack rather than open-ended AI-native web search.

FAQ

Asked in real client work.

What does the free tier actually include?

Exa's pricing page states up to 20,000 requests per month free, covering web search calls and webpage text/highlights retrieval. There's no card required to start; overage moves you to pay-as-you-go rates.

How is pay-as-you-go priced once I'm past free usage?

Search is $7 per 1,000 requests for up to 10 results; extra results beyond 10 cost $1 per 1,000 requests. Contents (page text) is $1 per 1,000 pages per content type. Deep Search runs $12-15 per 1,000 requests, Monitors $15 per 1,000, and the Answer endpoint starts at $5 per 1,000 requests.

Is there a separate cost for AI-generated summaries or agent features?

Yes. AI page summaries add $1 per 1,000 pages across endpoints. Exa's Agent product bills separately by compute unit at $0.10 per ACU, with fixed-effort modes from $0.012 to $1.00 per request; contact enrichment through the agent adds $0.02 per email found and $0.07 per phone number found.

Do startups get any pricing break?

Exa's site mentions a startup grant program offering $1,000 in free API credits for eligible startups, on top of the standard free tier.

What's the integration lift since this is API-only?

Because ease is listed as medium and interface is API, someone on the team needs to write the request/response handling and manage rate limits and error handling in code. There's no low-code connector shown on the vendor's own pages, so plan for developer time before Exa fits into a workflow like n8n or Make unless a community connector already exists.