I Compared the Major Web Search APIs Available on AWS Marketplace

Hello, I'm Keima.

I often hear requirements to add Web search functionality to AgentCore. When actually integrating it, you will need to choose some kind of Web search API, and when deciding which one to adopt, you can't ignore not only cost but also perspectives such as whether there are any security issues and whether it supports Zero Data Retention (ZDR).

So in this article, I take up four major services offered via AWS Marketplace among the representative Web Search APIs (Tavily Search / Brave Search / Perplexity Search API / Exa Search), and compare them by summarizing research results as of May 21, 2026. I have organized the information mainly around pricing models, data protection, and how to procure them via AWS Marketplace. I hope this serves as a reference for API selection.

Web Search API Candidates

• Tavily Search
• Brave Search
• Perplexity Search API
• Exa Search

Features of Each

Tavily Search

• Overview: A search and extraction API specialized for AI agents, which can return search results together with extracted body text at the same time.

• Pricing model: It adopts a simple credit-consumption model (Reference: Tavily Docs - Credits & Pricing).

  Price per plan and unit price per credit

  Plan name	Monthly fee	Monthly credit limit	Unit price per credit
  Researcher (Free)	$0	1,000	Free
  Project	$30	4,000	$0.0075
  Bootstrap	$100	15,000	$0.0067
  Startup	$220	38,000	$0.0058
  Growth	$500	100,000	$0.0050
  Pay as you go	Pay for what you use	No limit	$0.0080
  Enterprise	Custom	Custom	Custom

  Consumption per endpoint

  • Search: A basic search costs 1 credit per request, and advanced costs 2 credits.
  • Extract: For every 5 successful content extractions from URLs, basic costs 1 credit and advanced costs 2 credits. (*Failed URLs are not charged)
  • Map: 1 credit for every 10 pages successfully mapped from a URL. With instructions (natural language instructions) added, it costs 2 credits. (*Failures are not charged)
  • Crawl: The sum of the costs of the Map endpoint and the Extract endpoint.

• Safety (data protection):
  The standard plan explicitly states that "unless otherwise specified under the contract, certain portions of query data may be used to improve future queries," and it also officially acknowledges that queries may be shared with third-party search providers such as Google only when results cannot be retrieved from Tavily's own index.

  Additionally, unless otherwise specified under the contract between you and us, we may use certain portions of your query data to improve our responses to future queries.

  While we mostly use our own services, we may also share your query data with third-party search index providers in limited situations where our own search index is unable to retrieve the requested content in response to your query.

  Source: Privacy Policy | Tavily

  Independently audited for SOC 2 Type II

  Tavily’s systems and controls are independently audited under our SOC 2 Type II report, validating that our practices meet industry standards for security, confidentiality, privacy, and availability.

  Source: Trust Center | Tavily

• AWS/platform integration: The Tavily MCP Server can be subscribed to on AWS Marketplace and can be deployed as a managed MCP server on Amazon Bedrock AgentCore.

  Reference: Partnerships with Amazon | Tavily Docs

Brave Search

• Overview: A search API with its own independent index of over 30 billion pages.

• Pricing model: It adopts pay-as-you-go pricing (AWS Marketplace).

  • Price: $5 per 1,000 requests ($0.005 per request) on the Data for AI plan
  • Rate limit: Up to 50 queries per second
  • Monthly limit: Unlimited number of queries

• Safety (data protection):
  For the standard "Data for AI" plan and similar plans, the privacy policy explicitly states that query logs are retained for up to 90 days for billing and troubleshooting purposes.

  A record of search queries submitted to the Brave Search API via a customer’s Search API account is retained for a maximum of 90 days for the purpose of billing Search API account holders and for troubleshooting subject to Brave’s legal obligations.

  Source: Privacy Policy | Brave Search API

  In terms of compliance, it is SOC 2 Type II attested.

  The Brave Search API is SOC 2 Type II attested, with its strict security and privacy controls independently verified by a third-party auditor.

  Source: Security | Brave Search API

  "Zero Data Retention (ZDR)" is not included in the standard plan and is offered in the Enterprise plan. By upgrading your contract to Enterprise, you obtain the following features and benefits.

  • Full-funnel Zero Data Retention (complete zero data retention)
  • Custom contract and NDA execution
  • Support for custom capacity requirements
  • Invoice payment and enterprise-grade support

  Source: Plans | Brave Search API

• AWS/platform integration: Can be subscribed to via AWS Marketplace.

  Reference: Brave Search API | AWS Marketplace

Perplexity Search API

• Overview: An advanced AI-powered solution designed for building applications using real-time web knowledge and research capabilities.

• Pricing model: When procuring via AWS Marketplace, you can purchase from a minimum of $1,000 (1,000 credits), and credits are consumed from a pool shared by all endpoints (1 credit = 1 USD).

  A guide to how much you can use with 1,000 credits ($1,000):

  • Search API (pure search): $5.00 / 1,000 requests. There is no token-based billing; only the number of requests is charged.
    → Approximately 200,000 requests are possible

  Reference: AWS Marketplace | Perplexity API Docs

• Safety (data protection): It provides "Zero Data Retention (ZDR)" by default, which does not retain search prompts or response content and does not use them for training.

  By default, the API has zero day retention of user prompt data, which is never used for AI training.
  We do not retain any query data sent through the API and do not train on any of your data.

  In terms of information use, only billing metadata (number of requests, number of tokens, etc.) and account information (name, email, etc.) are collected, and users' query information is not used for training data or the like.
  The processing infrastructure is hosted in AWS North America regions.

  Our compute is hosted via Amazon Web Services in North America.

  Source: FAQ | Perplexity API Docs

  It also complies with security standards and certifications (SOC 2, HIPAA, CAIQlite, etc.).

  Perplexity is compliant with industry-leading security standards and certifications:

  • SOC 2 Type II Report
  • 2025 HIPAA Gap Assessment
  • CAIQlite

  Source: Privacy & Security | Perplexity API Docs

Exa Search

• Overview: A proprietary search engine designed for AI. It provides the following three core features.

  • Search: Search web pages using embedding-based (semantic) or keyword-based search
  • Contents: Retrieve clean, up-to-date parsed HTML content from the URLs of search results
  • FindSimilar: Search for and return pages semantically similar to a specified link

• Pricing model:
  On AWS Marketplace, it can be procured as an Enterprise plan (Exa Base Enterprise Plan) only as a "12-month contract only (1-year contract)." The list price is $48,000 / 12 months, and if the specified usage is exceeded, additional pay-as-you-go charges in $0.10 units are incurred.

  Reference: Exa Base Enterprise Plan | AWS Marketplace

• Safety (data protection): It provides Zero Data Retention (ZDR) by default.

  Today we’re excited to announce that Exa now offers Zero Data Retention (ZDR) for each of our search products -- search, answer, and deep research.

  ZDR is the golden standard for enterprises that require full data privacy. For a search provider to offer ZDR, it must never store user query data, neither in the main service nor any subprocessors.

  Source: Zero Data Retention across Exa Search Products | Exa Blog

  In terms of compliance, it is also SOC 2 Type II certified.

  Exa takes data security and privacy seriously. We are proud to be SOC 2 Type II certified, demonstrating our commitment to maintaining rigorous information security practices and controls.

  Source: Enterprise Documentation & Security | Exa Docs

Comparison Summary

BYOL (Bring Your Own License): One of the listing types on AWS Marketplace, a "bring your own license" model. Subscribing on AWS Marketplace does not incur charges, and a separate license contract and payment directly with the vendor are required. AWS Marketplace functions merely as a place to obtain activation rights for deploying to Amazon Bedrock AgentCore and the like, and usage fees are billed directly by the vendor.

PAYG (Pay As You Go): A pay-as-you-go pricing model where you are billed for what you use. There is no minimum contract amount or fixed monthly fee, and charges are based on the number of requests. When via AWS Marketplace, charges are consolidated into the AWS invoice.

Conclusion

When deciding which Web Search API to adopt, you need to make a comprehensive judgment that includes not only pricing but also security and data protection requirements. I hope this article serves as a starting point for selecting an API that fits each company's requirements.