Exa: what it does, pricing and alternatives

What it does

Exa is a search engine built for software rather than for people browsing a results page. Its homepage lists a Search API, a Contents API that returns page text, an Agent API for multi-step research with citations, a Monitors API that watches the web for changes, and Websets, a tool for building datasets. Websets is the part most relevant to building lists of leads, candidates or companies.

Exa's Websets FAQ describes the flow. A user writes a natural-language query and asks for a target number of results. Websets finds candidates and checks each one against the user's criteria, and only those that pass are kept. For each kept item it then runs enrichments: it searches for requested fields such as a CEO's name, a funding amount, an email address or a phone number. Results arrive in a dashboard, through polling, or by webhook. The FAQ notes that the requested count "is a target, not a guarantee." If too few candidates pass every rule, the list comes back shorter.

Who it is for

Two groups fit especially well. The first is people whose search criteria are too fuzzy for a filter-based contact database, such as "seed-stage climate startups that sell to utilities." The second is developers who want list-building and web search inside their own AI agents. The homepage says Exa has more than 500,000 developers and holds SOC 2 Type II, GDPR, CCPA and HIPAA (with a BAA) credentials for enterprise buyers.

How people use it

Take a recruiter or sales rep building a list of 100 heads of engineering at Series A fintech companies in Europe. They describe that in plain English, set the count to 100, and add criteria such as "company raised a Series A" and "person currently leads engineering." Websets searches, verifies each candidate against those rules, and fills in rows as they pass. The user then adds enrichment columns: an email, a phone number, and perhaps a research field like "most recent funding date." When the table is done, they export it. A competitor's comparison by Autotouch says documented exports include Salesforce, HubSpot, Instantly, Smartlead and Lemlist, plus Clay via webhook.

Autotouch's model of the published credit rates shows the cost. Discovering 100 matching people with email, phone and three research fields comes to about 2,600 credits. Enriching 100 person URLs the user already has comes to about 1,800. Autotouch also warns that a correct company does not prove a contact still holds the role, and a returned email does not prove it is deliverable. Both should be checked before outreach.

Pricing

Checked on 2026-10-21. According to Exa's pricing page, the API has a Free Tier: $10 in credits at signup (up to 2,500 Lite searches), a balance that resets to $10 every month, no payment method required, and a one-time $10 bonus after onboarding. After that, pay-as-you-go has no subscription or minimum spend. Search starts at $4 per 1,000 requests, Deep Search is $12 to $15 per 1,000, Contents is $1 per 1,000 pages, Answer is $5 per 1,000 and Monitors is $15 per 1,000. Agent runs cost from $0.012 to $1.00 each at fixed effort levels, or are metered, including $0.02 per email and $0.07 per phone number found. Enterprise pricing is custom.

Websets is billed in its own credits, separate from API credits. Its billing page returned a rate-limit error when we tried to open it. Search results quoting that page, along with Autotouch's analysis, report four tiers: a Free plan with 1,000 credits and Websets capped at 25 results; a Starter (also called Core) plan at $49 per month for 8,000 credits; Pro at $449 per month for 100,000 credits; and custom Enterprise. Matching results reportedly cost 10 credits each, emails and phone numbers 5 credits each, other enrichments 2 credits per row, and non-matching results nothing. The FAQ says the Websets API requires a paid plan. Since the tier names differ between Exa's own pages, check the live billing page before buying.

Strengths

The main draw is expressing a list in plain language instead of fixed database filters, with a verification step that drops results that fail the stated criteria. The one independent benchmark found, a March 2026 NewsCatcher test of 32 event-discovery queries cited on a competitor's site, gave Websets 19.6% recall and 83.7% precision, against 5.5% and 77.7% for Parallel FindAll Core. The same platform also offers a cheap, usage-based API, so a developer can prototype with free credits and move to an agent or monitor later. Product Hunt reviews from founders praise semantic relevance and easy integration, though all visible reviews there are positive and come from builders.

Limits and things to watch

The most detailed public criticism appears in the May 2025 Launch HN thread, "Exa (YC S21) – The web as a database." Commenters reported irrelevant matches, including a GitHub search that one user put at roughly 30% on target. Others said a basic OR criterion was misunderstood and city-based searches returned far-away companies. Several described slowness: one search took about 15 minutes to verify the first row, and another user called it too slow for daily use. Some saw stalled UIs and results stuck on "Verifying...". On cost, one commenter said a single four-criteria search used 750 of their 1,000 free credits, and another declined to pay $49 a month for the speed and results they saw. That feedback is from launch, and the product may have improved, but no newer independent review was found to confirm it.

Autotouch, which sells a competing tool, points to structural limits. Starter caps a Webset at 100 results and 10 enrichments. Exports do not provide continuous two-way CRM sync. And Exa now steers new list-building integrations toward its Agent product. The FAQ itself recommends Agent Ultra for lists of 1,000 or more items, so heavy users may end up with two meters, Websets credits and API spend.

When to choose something else

Clay suits teams running repeatable outbound programs. It has a $0 plan with 500 Actions and 100 Data Credits a month, and paid plans starting with Launch at $167 a month on annual billing ($185 monthly), with Actions sold in separate volume tiers. It also offers multi-provider enrichment waterfalls and unlimited seats. Parallel FindAll is the closer API rival: there is no subscription, and each query costs a fixed fee plus a per-match fee (Core is $2.00 plus $0.15 per match), with citations and confidence scores on each condition. Pipeline builders who dislike monthly credits may prefer it, though the NewsCatcher benchmark favored Exa. Apollo.io is the better fit when a known title, industry and company size are enough, because filtering an existing contact database and sequencing emails in the same tool is simpler. Its pricing page describes Free, Basic, Professional and Custom plans, and Capterra lists Basic at $49 per user per month.

Verdict

Exa is worth trying for anyone whose ideal list is easier to describe in a sentence than to build from dropdown filters. That includes investors hunting niche companies, recruiters with unusual profiles, and developers wiring list-building into an agent. The free credits make testing cheap. Credits do run out quickly, launch-era users reported slowness and misses, and contact details still need verification. Teams that depend on high-volume, CRM-connected prospecting at a predictable price per contact are better served today by Clay or Apollo, and should revisit Exa when more independent results are published.