Amazon Kendra: What It Is and When to Use It

Definition

Amazon Kendra is a highly accurate and intelligent enterprise search service powered by machine learning (ML). It solves the common business problem of information silos by enabling users to find precise answers from vast amounts of unstructured and structured data across disparate repositories using natural language queries. Instead of returning a simple list of documents, Kendra provides direct answers, document excerpts, and FAQs, fundamentally changing the search experience from "search and scroll" to "ask and answer".

How It Works

Amazon Kendra's architecture is designed to be a fully managed service, abstracting away the complexity of building and maintaining a sophisticated search platform. The process flows from data ingestion to query resolution.

Create an Index: The core component of Kendra is the index, a container for your searchable data. When you create an index, you choose an edition (Developer or Enterprise) which determines its scale and availability.
Add Data Sources: You populate the index by connecting to your existing data repositories using built-in connectors. Kendra offers a wide range of native connectors for sources like Amazon S3, SharePoint, Salesforce, Confluence, Google Drive, and relational databases (via Amazon RDS). You can also add documents directly using the BatchPutDocument API or add question-answer pairs via an FAQ file.
Sync and Ingest: Once a data source is configured, you initiate a sync job. Kendra crawls the source, extracts text from various document formats (PDF, HTML, Word, PowerPoint, etc.), and ingests the content into the index. You can schedule these syncs to run periodically to keep the index up-to-date with any changes in the source repository.
ML-Powered Enrichment and Indexing: This is where Kendra's intelligence shines. During ingestion, it doesn't just perform keyword indexing. It uses deep learning models to understand the content, its structure, and the relationships between concepts. It can recognize document titles, excerpts, and FAQ pairs. For more advanced use cases, Custom Document Enrichment allows you to use AWS Lambda functions to pre-process documents—for example, to perform Optical Character Recognition (OCR) on scanned PDFs or to add custom metadata before indexing.
Querying: Users and applications interact with the index via the Query API, submitting questions in natural language. Kendra's query engine processes the question, understands the user's intent, and searches the index. It returns several types of results:
- Suggested Answers: Direct, machine-generated answers extracted from the text.
- FAQ Matches: Exact matches from an uploaded FAQ file.
- Document Matches: A ranked list of the most relevant documents, with highlighted excerpts showing where the answer might be found.
Relevance Tuning: Kendra's ML models handle relevance ranking automatically. However, you can influence the results by boosting certain attributes, such as document freshness or specific data sources, to align the search experience with business priorities.

Key Features and Limits

Key Features

High-Accuracy Semantic Search: Uses Natural Language Processing (NLP) to understand the context and intent of a query, delivering more relevant results than traditional keyword search.
Native Connectors: Simplifies data ingestion from over 40 common data sources, including Amazon S3, Microsoft SharePoint, Salesforce, and Google Drive.
RAG Retriever API: Provides an optimized API specifically for Retrieval-Augmented Generation (RAG) use cases, feeding LLMs with semantically relevant and smartly chunked passages to improve generative AI application accuracy.
Custom Document Enrichment: Allows pre-processing of documents with AWS Lambda functions during ingestion for tasks like OCR, entity extraction, or metadata transformation.
Experience Builder: A no-code, drag-and-drop visual tool to quickly build and deploy a fully functional and secure search application, complete with user authentication via AWS IAM Identity Center.
Relevance Tuning: Administrators can boost the importance of specific fields, data sources, or document freshness to fine-tune search results.
Security and Access Control: Integrates with identity providers and respects document-level Access Control Lists (ACLs), ensuring users only see results they are permitted to see.

Service Limits (Quotas)

Indexes per Account: 5 Enterprise Edition and 5 Developer Edition indexes per account (adjustable).
Data Sources per Index: 50 for Enterprise Edition, 5 for Developer Edition.
Document Size: Maximum file size of 50 MB, with a maximum extracted text size of 10 MB.
Queries per Second (QPS): Each query unit supports approximately 0.1 QPS (about 8,000 queries per day). Enterprise Edition indexes can scale up by adding more query units.

Note: Service quotas can often be increased by contacting AWS Support.

Common Use Cases

Internal Knowledge Base Search: Empower employees to quickly find information in internal documentation, HR policies, technical wikis, and project repositories, reducing time spent searching and improving productivity.
Enhanced Customer Support: Power customer-facing websites, self-service portals, and chatbots with a high-accuracy search engine that can provide direct answers from knowledge bases and FAQs, deflecting support tickets.
Research and Development Acceleration: Enable scientists, engineers, and researchers to search through vast archives of research papers, patents, and experimental results to accelerate innovation and avoid redundant work.
Website Search Modernization: Replace legacy keyword-based search on public-facing websites with an intelligent, Google-like experience that helps users find what they need faster.

Pricing Model

Amazon Kendra follows a pay-as-you-go model with no upfront costs, billed on an hourly basis. The pricing structure has several components:

Index Edition: You are charged an hourly rate for running a Kendra index, which varies by edition.
- GenAI Enterprise Edition: The highest accuracy edition, optimized for RAG workloads.
- Basic Enterprise Edition: A high-availability option for production workloads.
- Basic Developer Edition: A lower-cost option for proof-of-concepts and development, not recommended for production.
Storage and Query Units: The base hourly price for an index includes a certain amount of document storage and query capacity. You can add capacity by purchasing additional storage units and query units.
Connector Usage: You are charged an hourly rate for each connector that is actively running and syncing data.

There is a 30-day free trial for new customers that includes 750 hours of usage for the Basic Developer Edition or GenAI Enterprise Edition. For detailed and up-to-date pricing, always consult the official Amazon Kendra Pricing page and the AWS Pricing Calculator.

Pros and Cons

Pros

High Accuracy Out-of-the-Box: The primary strength is its ML-powered semantic understanding, which provides highly relevant answers to natural language questions with minimal configuration.
Fully Managed Service: AWS handles all the underlying infrastructure, scaling, and maintenance, allowing developers to focus on building the application.
Broad Connectivity: An extensive library of built-in connectors makes it easy to index data from many popular enterprise systems.
Strong Security: Integrates with enterprise identity systems and enforces document-level permissions, ensuring data governance.

Cons

Cost: Kendra is a premium service and can be significantly more expensive than self-managed alternatives like Amazon OpenSearch Service, especially for large-scale deployments.
Indexing Latency: Syncing data sources and indexing large volumes of documents can take a considerable amount of time.
Configuration Complexity: While basic setup is straightforward, advanced features like relevance tuning and custom enrichment can have a steep learning curve.
"Black Box" Nature: As a managed service, you have less control over the underlying ranking algorithms and vectorization models compared to building a solution on OpenSearch or with vector databases.

Comparison with Alternatives

Amazon Kendra vs. Amazon OpenSearch Service

Primary Use Case: Kendra is an answer engine optimized for natural language, question-answering, and enterprise document search. OpenSearch is a more general-purpose search and analytics engine, ideal for log analytics, real-time monitoring, and customizable full-text search.
Management & Expertise: Kendra is fully managed and requires no ML expertise. OpenSearch requires you to manage clusters, define schemas (mappings), and actively tune relevance, demanding more technical expertise.
Query Type: Kendra excels at semantic, natural language queries ("How do I reset my password?"). OpenSearch is powerful for keyword, faceted, and complex structured queries but requires significant effort to achieve similar semantic capabilities.

Amazon Kendra vs. Knowledge Bases for Amazon Bedrock

Role and Scope: Kendra is a full-fledged, standalone enterprise search service with features like a UI builder, faceting, and connectors. A Bedrock Knowledge Base is a component within the Bedrock service designed specifically to provide data for Retrieval-Augmented Generation (RAG) to a Large Language Model (LLM).
Function: The primary function of a Bedrock Knowledge Base is to retrieve contextual information to feed to an LLM to generate a conversational answer. The primary function of Kendra is to find the answer itself within the documents.
Integration: The two are not mutually exclusive. You can use an Amazon Kendra index as the underlying retriever for a Bedrock Knowledge Base, combining Kendra's powerful search and connector capabilities with Bedrock's generative AI.

Exam Relevance

Amazon Kendra is a relevant topic in several AWS certification exams, particularly those focused on solutions architecture and machine learning.

AWS Certified Solutions Architect - Associate (SAA-C03): You should understand Kendra's core use case as an intelligent enterprise search service and be able to differentiate it from Amazon OpenSearch Service. Know when to choose Kendra for building a corporate knowledge base or website search.
AWS Certified Developer - Associate (DVA-C02): Be familiar with how an application would integrate with Kendra at a high level, primarily through its Query API to submit user questions and receive results.
AWS Certified Machine Learning - Specialty (MLS-C01): Expect deeper questions. Understand its architecture, the role of data source connectors, the concept of semantic search vs. keyword search, and how it fits into a broader AI/ML application, including its role in RAG pipelines.

Frequently Asked Questions

Q: How is Amazon Kendra different from Amazon OpenSearch Service?

A: The primary difference is their purpose. Amazon Kendra is a fully managed, ML-powered service designed to provide direct answers from unstructured documents using natural language. Amazon OpenSearch Service is a highly customizable, self-managed search and analytics engine best suited for log analysis, monitoring, and complex keyword-based searches where you need fine-grained control over the infrastructure and relevance algorithms.

Q: Can Amazon Kendra search both unstructured and structured data?

A: Yes. Kendra excels at searching unstructured text within documents like PDFs, Word files, and HTML pages. It also leverages structured data, such as document metadata (author, creation date, category) and database fields, to allow for powerful filtering and faceting of search results.

Q: What is the difference between Amazon Kendra and a Knowledge Base for Amazon Bedrock?

A: Amazon Kendra is a complete enterprise search service designed to find answers within your documents. A Knowledge Base for Amazon Bedrock is a feature that connects your data sources to a large language model (LLM) to provide context for generating new answers (a process called RAG). You can use a Kendra index as the search component within a Bedrock Knowledge Base, leveraging the strengths of both services.

This article reflects AWS features and pricing as of 2026. AWS services evolve rapidly — always verify against the official AWS documentation before making production decisions.