Exa review

Exa is a hosted search and web-crawling service exposed through APIs, not a search library or self-hosted server. It sits between an AI agent and web or private-information retrieval: the application sends a query, then uses results, page content or structured research in its own context and workflow.

The Search API offers configurable latency and coverage across web, news, companies, research, people and financials. The Contents API returns webpage text and relevance-selected highlights. Agent supports research runs with structured outputs and grounded citations, while Monitors runs searches on a cadence and can send updates through webhooks. Exa also offers structured company enrichment across 70M+ companies.

That makes Exa a retrieval layer rather than a complete agent runtime. Your application still owns orchestration, user-facing behavior and handling returned data. Exa says its index searches public web and private information; teams using private sources should confirm how those sources are connected and governed before relying on that capability.

Build around Exa’s Search, Contents, Agent and Monitors APIs. Search is configurable for latency and coverage; Contents can return relevant excerpts instead of requiring an application to pass full pages into a model. Agent is for deeper research with structured outputs and citations. Monitors supports recurring searches and webhook updates. Exa Connect is another listed integration surface.

The homepage points developers to Docs and an API Playground, and lists 50+ integrations on the Free plan. The vendor does not name SDK languages or a CLI on its homepage, so confirm current SDK support and API request conventions in the docs before choosing an implementation. A Reddit developer describes using Exa through an MCP server, but that is a community integration, not a substitute for checking Exa’s supported interfaces.

Keep the integration boundary narrow: normalize Exa responses in your application and preserve the citations or URLs your product needs. For long-running research and monitoring, also design for asynchronous completion and webhook delivery behavior; confirm those operational details in the docs.

The Free plan costs $0 and includes $10 in monthly credits plus a $10 onboarding bonus, 10 Search QPS and 50 Agent concurrency. Pay as you go is usage-based, supports up to 25 Search QPS and includes 50 Agent concurrency. Custom lists 25 Search QPS, custom Search QPS and Agent concurrency, and up to 1,000 results per search. The pricing page lists no annual subscription rates for Pay as you go.

Search add-ons cost $4 per 1,000 Lite requests, $7 for Auto, $12 for Deep and $5 for Answer. Contents costs $1 per 1,000 pages; Monitors costs $2 per 1,000 requests. For Lite, Auto, Deep and Monitors, additional results above 10 cost $1 per 1,000 requests. AI page summaries are $1 per 1,000 pages.

Agent effort ranges from $0.012 per Minimal request to $1.00 per X-high request; Ultra is usage-based. Agent usage also lists $0.10 per ACU, $0.005 per search, $0.02 per email enrichment and $0.07 per phone enrichment. Exa Connect provider usage is billed separately. Custom zero-data-retention terms list a $12,000 annual commitment in API credits.

Exa is presented as a hosted API service. The vendor does not publish a source licence or self-hosted deployment option on its product homepage, so treat Exa as hosted-only unless the vendor confirms otherwise for your contract. That makes the service quick to add as a retrieval dependency, but it also ties your application to Exa’s API behavior, pricing and availability.

The vendor says Custom includes a zero-data-retention option; the pricing page lists a $12,000 annual commitment in API credits for the enterprise option. That is a retention control, not a data-export guarantee. The vendor does not describe a general export path for queries or results. Keep needed source URLs, citations and normalized outputs in your own storage, and ask about data deletion, export and contract-end handling before sending sensitive workloads.

Migration means replacing search, content retrieval and any agent or monitoring workflows that depend on Exa-specific outputs. A small adapter layer can reduce that work.

The pricing page sets the Free plan at 10 Search QPS and Pay as you go at up to 25 Search QPS; Custom lists 25 Search QPS plus custom capacity. Agent concurrency is 50 on Free and Pay as you go, with custom concurrency available on Custom. Confirm whether these limits fit peak traffic and what happens when a request exceeds them. The vendor’s homepage does not publish an uptime commitment, so ask about service-level terms, retries and support response times.

Exa lists SOC 2 Type II, GDPR, CCPA and HIPAA safeguards, with a BAA available. Zero data retention is an enterprise option, not a general Free-plan promise. Verify the exact retention and processing terms for your account.

Reddit reports are mixed and anecdotal. One OpenCode user said search queries had started failing, then edited the post to say the issue was resolved. A Pi developer describes a community MCP integration with 1,000 requests per month and no API key; a commenter warns that multiple search agents can use up quota quickly. Budget from your own request pattern, not a headline quota.