Amazon Kendra Guide: Features, Pricing and the 2026 Change

What Is Amazon Kendra?

McKinsey research found that knowledge workers spend about 1.8 hours a day searching for and gathering information, moving between disconnected systems, scanning keyword results and digging through document stores that keep growing. Keyword search often fails because users do not know the exact terms a document uses. Someone types “what is our vacation policy” and gets nothing, because the HR document says “paid time off”. Amazon Kendra was built to solve this with ML-powered natural language search. Before you plan around it, though, note the 2026 change: AWS put Amazon Kendra into maintenance mode on June 30, 2026, and closed it to new customers on July 30, 2026.

Amazon Kendra is a fully managed enterprise search service from Amazon Web Services, powered by machine learning. It uses natural language processing to work out the intent behind a query and return relevant answers, instead of only a ranked list of documents that might match. So users can ask plain questions (“How long is maternity leave?” or “How do I configure my VPN?”), and Kendra returns answers with confidence scores and source citations.

Amazon Kendra connects to existing data sources, including S3, SharePoint, Confluence, Salesforce, ServiceNow, OneDrive and Google Drive, through more than 30 built-in connectors. In addition, these crawl and index content on a schedule and keep the access controls from each source system. As a result, employees only see results from documents they are allowed to open, without a separate permission layer.

Kendra Enterprise Impact

Amazon Kendra also became a retrieval layer for retrieval-augmented generation (RAG). Its Retrieve API returns relevant passages that can be passed to foundation models on Amazon Bedrock, so answers are grounded in your own enterprise content and hallucinations are reduced. For new projects, however, AWS now recommends Amazon Bedrock Managed Knowledge Base, a fully managed RAG service with built-in connectors, hybrid search and a RetrieveAndGenerate API.

Supported Document Formats

Kendra supports many document formats, including PDF, HTML, Word, PowerPoint, plain text and FAQ lists in CSV. It extracts text, structure and metadata automatically, without custom parsing or conversion pipelines. So organizations can index formal policies, wiki pages and customer-facing knowledge articles through one permission-aware search interface.

Kendra also works with other AWS services: Amazon Lex for chatbot interfaces, Amazon Comprehend for text analysis and AWS Lambda for custom processing. Existing customers can therefore run a complete knowledge system inside AWS, from indexing through natural language search to AI-generated answers.

How Amazon Kendra Works

Amazon Kendra indexes content from connected data sources, then applies machine learning models to understand the meaning of each query and match it to the most relevant information in the index.

Kendra Index Architecture

Every Kendra deployment is built around an index, a searchable store of your enterprise content. You create an index and connect data sources, and Kendra crawls, processes and indexes the documents. The index stores the raw text along with metadata, document structure, access control lists and semantic representations that support intelligent retrieval.

During indexing, Kendra applies several ML steps to each document: text extraction from PDF, HTML, PowerPoint, Word and other formats, entity recognition, key phrase identification and semantic representation. That lets Kendra match on meaning as well as keywords. For example, “Who approves overtime?” can match a document titled “Management Authorization Procedures”, even though the two share no keywords.

Amazon Kendra offers two classic index editions for different scales:

• Developer Edition: Supports up to 10,000 documents and 4,000 queries per day. It is meant for development, testing and small deployments, and runs in a single availability zone.
• Enterprise Edition: Supports up to 100,000 documents and 8,000 queries per day in its base configuration, with more capacity available through additional storage and query units. It is built for production and runs across three availability zones.

AWS also introduced the GenAI Index, an index type designed for retrieval-augmented generation and intelligent search. It uses newer semantic models and retrieval techniques to help enterprises build digital assistants and search experiences, and it is tuned for the Kendra Retrieve API that feeds results to foundation models in Bedrock.

Three Types of Kendra Results

Where traditional search engines return a ranked list of documents, Amazon Kendra returns three types of results for each query, each giving a different level of detail:

• Factoid answers: Direct answers extracted from documents with high confidence. For “How long is maternity leave?”, Kendra might return “14 weeks”, taken from your HR policy with the source cited. This is the most useful result type, because the user gets the answer without opening a document.
• Document excerpts: Relevant passages with the answer highlighted in context. For broader questions such as “How do I configure my VPN?”, Kendra extracts the most relevant paragraphs, so the user gets enough context without reading the whole document.
• Document rankings: A ranked list of relevant documents, used as a fallback when no specific answer can be extracted. Users still get useful results for ambiguous or specialized queries.

So Kendra gives precise answers when it can, excerpts when the question is broad and document links when neither fits. As a result, users open fewer documents than with keyword search. Each result also carries a confidence score, which your application can use to show factoid answers prominently when confidence is high and fall back to a document list when the query is ambiguous.

Data Source Connectors for Kendra

Amazon Kendra provides more than 30 native data source connectors that index content where it already lives, with no data migration:

• Cloud storage: Amazon S3, OneDrive, Google Drive, Box
• Collaboration platforms: SharePoint, Confluence, Slack
• Business applications: Salesforce, ServiceNow, Jira
• Databases: Amazon RDS, custom JDBC-compatible databases
• Web content: Web crawler for indexing website content
• Custom sources: Custom connector SDK for proprietary data sources

Each connector syncs access control lists (ACLs) from the source system. When a user searches, Kendra filters results by that user’s identity and group memberships, so they only see documents they are authorized to open. As a result, you do not need to rebuild permission models, and results stay in line with your existing access policies.

Core Amazon Kendra Features

Beyond the index and connectors, several features make Amazon Kendra suited to enterprise use. Together they let search improve over time, learning from user behavior, adapting to your vocabulary and giving you control over result relevance:

Kendra and Generative AI

In recent years, the biggest change to Amazon Kendra was its role as a retrieval layer for generative AI. The Kendra and Bedrock RAG pattern became a common architecture for enterprise AI assistants, so existing customers should understand how it works, and how it maps to Bedrock Managed Knowledge Base if they migrate.

Kendra Retrieve API for RAG Workflows

Kendra’s Retrieve API is built for RAG workflows. The standard Query API returns document excerpts, while the Retrieve API returns passage-level results sized to improve what is sent to the foundation model. In turn, that removes the need to build custom chunking, embedding and retrieval logic, which is one of the harder and slower parts of a RAG project.

The RAG workflow runs like this:

1. A user asks a natural language question through your application interface.
2. The Kendra Retrieve API semantically searches your enterprise content and retrieves the most relevant document passages.
3. These passages are assembled into a context payload and sent to a foundation model on Amazon Bedrock (such as Anthropic Claude, Meta Llama, or Amazon Nova).
4. The foundation model generates a natural language answer grounded in the retrieved passages, citing specific sources.
5. The answer is returned to the user with source citations from your original documents.

This design combines the fluency of a large language model with answers grounded in your enterprise documents, which reduces hallucination risk compared with letting a model answer from its training data alone.

Access Control in Kendra RAG Workflows

Kendra also applies document-level access controls, so RAG answers only draw on documents the requesting user is allowed to see. That is a real advantage over generic vector databases without built-in access control. The same ACL sync that governs direct search also governs RAG retrieval, which helps keep AI answers from exposing restricted documents, a requirement in regulated industries.

The GenAI Index was built for this RAG pattern. It uses newer semantic models and retrieval techniques to improve retrieval accuracy over the standard index, at the passage level foundation models need. Existing customers running assistants on Kendra can keep this stack, while AWS points new builds to Bedrock Managed Knowledge Base.

Talk to an ExpertRunning Amazon Kendra? Plan your move to Bedrock Knowledge Bases with us.

Amazon Kendra Pricing Model

Unlike per-request services, Unlike per-request services, Amazon Kendra uses subscription-based pricing with monthly fees for index capacity. Rather than listing specific dollar amounts that change over time, here is how the cost structure works:

Understanding Kendra Cost Dimensions

• Index hourly rate: You pay for each hour the index is running. The rate differs between Developer and Enterprise editions, and Enterprise costs more because it adds multi-AZ availability, higher document limits and more query capacity.
• Document storage units: Capacity beyond the base document limit comes in storage units. Each unit raises the number of documents you can index and adds to the monthly cost.
• Query capacity units: Throughput beyond the base query limit comes in query capacity units. Similarly, each unit raises the number of queries per day.

Connector and Free Tier Costs

• Connector sync: Connectors run on scheduled sync intervals. More frequent syncs mean higher connector costs, although these are usually small next to index charges.
• Free tier: Developer Edition included 750 free hours in the first 30 days for evaluation. Since Kendra is closed to new customers, new accounts can no longer use it, and Enterprise Edition never had a free tier.

Amazon Kendra Security and Compliance

Kendra indexes some of your most sensitive content, such as HR policies, financial reports, legal contracts, customer data and technical documentation, so security matters in every production deployment.

Index data is encrypted at rest with AWS KMS, including support for customer-managed keys, so you control the encryption keys. API traffic is encrypted in transit with TLS. Kendra also supports interface VPC endpoints through AWS PrivateLink, and connectors can reach data sources inside your VPC, so traffic between your network and Kendra can stay off the public internet.

Amazon Kendra is covered by AWS compliance programs, including SOC, ISO and PCI DSS, and it is HIPAA eligible. Check AWS Services in Scope for the current list in your region. The ACL sync from connected sources means results respect your existing access policies without duplicate permission management.

Kendra filters results at the document level by user identity and group membership. When a user searches, Kendra shows only documents that user may access, even across several data sources with different permission models. That built-in access control is what sets Kendra apart from simply feeding documents into a generic search engine or LLM.

All Kendra API calls are logged in AWS CloudTrail, which gives you an audit trail for compliance reporting and security investigations. Together with Kendra’s query logs, organizations can see who searched, when queries ran and which results were returned, which helps show compliance with data access policies.

What’s New in Amazon Kendra

Amazon Kendra received several important updates before it moved to maintenance mode, with generative AI integration and the GenAI Index the biggest changes:

So Amazon Kendra grew from a standalone search service into a retrieval layer for AWS enterprise AI, used through direct search, Bedrock RAG, Q Business assistants and Lex chatbots. In maintenance mode, AWS still fixes bugs and ships security updates for existing customers, but it builds no new features. Teams running Kendra should plan their next step, either staying on Kendra for stable workloads or migrating to Bedrock Managed Knowledge Base.

Real-World Amazon Kendra Use Cases

With its natural language understanding and wide connector support, Amazon Kendra serves industries where finding information quickly affects productivity, compliance or customer satisfaction. These are the most common use cases among existing customers.

Most Common Kendra Implementations

Below are the use cases we see most often in client environments:

Amazon Kendra vs Azure AI Search

If you are comparing enterprise search services across cloud providers, Amazon Kendra and Azure AI Search take two quite different approaches to the same problem. Here is how they compare on the points that matter most:

Choosing the Right Kendra Alternative

Both are mature enterprise search platforms with different priorities. Amazon Kendra favors fast deployment with little operational work: you connect data sources, and Kendra provides natural language Q&A with automatic access control and incremental learning. Azure AI Search, by contrast, offers deeper customization through AI enrichment pipelines, native vector search and explicit scoring profiles, which suits teams that want fine control over search behavior.

Your cloud ecosystem largely decides the fit. Kendra is now closed to new customers, so new AWS projects should compare Azure AI Search with Bedrock Managed Knowledge Base. Teams on Azure with Azure OpenAI will find that Azure AI Search integrates natively with the Azure AI stack.

For AWS teams that need full-text and vector search at a lower baseline cost, Amazon OpenSearch Serverless is an option, although it lacks Kendra’s built-in natural language understanding, automatic ACL sync and incremental learning. Many enterprises combine Kendra for knowledge Q&A about policies, procedures and documentation with OpenSearch for high-volume application search, such as product catalogs, log analytics and real-time data.

Kendra vs Bedrock Knowledge Bases

Amazon Bedrock Knowledge Bases is AWS’s recommended replacement. AWS describes Bedrock Managed Knowledge Base as a fully managed RAG service with built-in connectors, smart parsing, hybrid search, a RetrieveAndGenerate API and agentic retrieval across several knowledge bases. Some Kendra features have no direct equivalent yet, including query suggestions, faceted search, custom synonyms, spell checking, incremental learning and custom document enrichment, and the managed service has 7 connectors against Kendra’s 32+. AWS’s migration guide gives workarounds for each gap.

Getting Started with Amazon Kendra

Amazon Kendra is no longer open to new customers, so this section is for existing customers adding indexes or applications. Setup stays simple: you create an index, connect data sources and start searching, with no ML expertise needed.

Setting Up Your First Kendra Index

In the Amazon Kendra console, create a new index and pick the edition that fits your scale. Then configure a data source connector, starting with S3 if your documents are already in AWS, or SharePoint and Confluence for collaboration content. Kendra then crawls and indexes the connected content.

Here is a minimal Python example that queries an existing Kendra index:

import boto3

# Initialize the Kendra client
client = boto3.client('kendra', region_name='us-east-1')

# Query the index
response = client.query(
    IndexId='YOUR_INDEX_ID',
    QueryText='How long is maternity leave?'
)

# Print results by type
for result in response['ResultItems']:
    print(f"Type: {result['Type']}")
    print(f"Answer: {result.get('DocumentExcerpt', {}).get('Text', 'N/A')}")
    print(f"Source: {result.get('DocumentTitle', {}).get('Text', 'N/A')}")
    print(f"Confidence: {result.get('ScoreAttributes', {}).get('ScoreConfidence', 'N/A')}")
    print("---")

For RAG, use the Kendra Retrieve API to pass retrieved passages to Amazon Bedrock. To add conversational search, connect Kendra to Amazon Lex bots through the built-in Kendra search intent. For setup details, see the Amazon Kendra documentation.

Amazon Kendra Best Practices and Pitfalls

From our work with Kendra in enterprise environments, a few patterns decide whether a deployment gives employees search they prefer over their old tools, or disappoints them with irrelevant results. The difference usually comes from how you configure, tune and feed it.

Recommendations for Kendra Deployment

• Start with your most-searched document repository: Connect the knowledge base that generates the most internal searches, usually HR policies, IT runbooks or product documentation. Proving value there builds support for a wider rollout.
• Invest in document metadata: Add metadata attributes such as department, document type, author and last updated date during indexing. Metadata enables relevance tuning, faceted filtering and more precise results, so its quality directly shapes search quality.
• Upload curated FAQ lists: Kendra’s FAQ matching model is accurate for structured question-and-answer pairs. Upload your existing FAQ content in CSV format to give instant, high-confidence answers to the most common employee questions; this is often the single most effective step for early search quality.
• Define custom synonyms for your organization: Map internal terms to everyday language. If your company says “PTO” but employees search for “vacation days”, synonyms close that gap across all queries without changing documents.
• Plan your path to Bedrock Managed Knowledge Base: Kendra gets bug fixes and security updates, but no new features. Run Kendra and Bedrock Managed Knowledge Base side by side on the same queries, compare relevance and latency, and decide which workloads to migrate and which to keep on Kendra.

When Not to Use Kendra

Kendra was built for enterprise document search, and it is the wrong tool for several common needs. For e-commerce product search with facets, price ranges and stock levels, use Amazon OpenSearch or Amazon Personalize. For real-time log analytics and monitoring, use Amazon OpenSearch or CloudWatch. And for structured data queries in SQL, use Amazon Athena or Amazon RDS. In short, Kendra is strong at “What does our policy say about X?” and weak at “Show me all products under $50 in electronics”.

Our ServiceData Analytics, Business Intelligence and Artificial Intelligence

Frequently Asked Questions About Amazon Kendra

References

1. Maintenance Mode June 30, 2026; Closed to New Customers July 30, 2026; 32+ Connectors: AWS — Amazon Kendra Availability Change
2. Kendra and Amazon Q Business Moving to Maintenance: AWS What’s New — AWS Service Availability Updates, June 2026
3. Index Editions, Result Types and Connectors: AWS — Amazon Kendra Developer Guide
4. 1.8 Hours a Day Searching for Information: McKinsey Global Institute — The Social Economy