AWS AI & Machine Learning
Kendra
Enterprise intelligent search service with natural language understanding
Amazon Kendra is a fully managed intelligent enterprise search service powered by machine learning that understands natural language queries and returns precise answers extracted from your documents - not just a list of links. It indexes content from S3, SharePoint, Confluence, Salesforce, databases, and custom sources, and uses semantic understanding to answer questions like "What is the IT security policy for remote work?" directly. For cloud engineers, Kendra fills the gap between keyword search (OpenSearch) and conversational AI (Bedrock RAG) for internal knowledge management.
Kendra vs OpenSearch - Choosing the Right Search Tool
Kendra and OpenSearch both do search but for very different use cases. This comparison comes up frequently in architecture discussions.
| Aspect | Amazon Kendra | Amazon OpenSearch |
|---|---|---|
| Query model | Natural language ("What is our vacation policy?") | Keyword/query DSL ("vacation AND policy") |
| Response type | Precise answer extracted from document + source | Ranked list of matching documents |
| Training required | None - semantic model pre-trained | None for keyword; ML features need training |
| Data sources | 40+ native connectors (SharePoint, S3, Confluence) | Manual ingestion or custom pipelines |
| Best for | Internal knowledge bases, HR/IT self-service portals | Log analytics, e-commerce search, custom ranking |
| Pricing model | Fixed per-edition pricing (not per query) | Per-AZU hour + storage |
| Maintenance | Fully managed, no tuning required | You manage index settings, mappings, analyzers |
Kendra is commonly paired with Amazon Q for Business. Kendra provides the search index layer; Q for Business adds a conversational AI interface on top. Many organizations replace standalone Kendra deployments with Q for Business.
Data Sources and Connectors
Kendra's native connectors automatically crawl, index, and sync content from enterprise systems. Incremental sync keeps the index current without full re-indexing.
| Source Category | Connectors Available |
|---|---|
| AWS Storage | S3, RDS, Aurora |
| Microsoft | SharePoint Online/On-premises, OneDrive, Teams, Exchange |
| Atlassian | Confluence Cloud/Server, Jira Cloud/Server |
| Salesforce | Salesforce Knowledge, Chatter, Cases |
| Custom | Custom Data Source (API for any system) |
| Web | Web Crawler (public URLs with configurable depth) |
| Database | JDBC sources via Custom Data Source |
Synchronization schedules can run on-demand or on a schedule (every 1 hour, 6 hours, or daily). New documents are typically searchable within minutes of being indexed.
Kendra Answer Types and Features
Kendra returns three types of results depending on query confidence - direct answers, featured snippets, and document links.
| Result Type | When Returned | Example |
|---|---|---|
| Direct Answer (FAQ) | Query matches a Q&A pair exactly or semantically | Q: "What is PTO policy?" -> A: "Employees receive 20 days PTO per year." |
| Excerpt/Snippet | Relevant passage found in document | Highlighted paragraph from HR handbook PDF |
| Document Result | Matching document found but no precise passage | Link to SharePoint document + metadata |
Faceted search, document enrichment, custom attributes, and access control lists (ACLs) let you filter results by metadata (department, document type, date) and enforce who can see which documents based on their identity provider groups.
Kendra ACL (access control) for document-level security requires you to configure user context in every query. If you pass a user token, Kendra filters results to only documents that user has access to. If you do NOT pass user context, all users see all indexed documents - a potential data exposure risk in multi-tenant deployments.
Kendra Pricing Editions
| Edition | Price | Documents | Queries/Day | Data Sources | Best For |
|---|---|---|---|---|---|
| Developer Edition | $810/month | 10,000 docs (5 GB) | 4,000 queries | 5 connectors | Proof of concept, dev/test |
| Enterprise Edition | $1,008/month base | 100,000 docs (30 GB) included | 40,000 queries | Unlimited connectors | Production enterprise search |
| Enterprise - extra docs | $0.002/doc/month beyond base | Up to 1M documents | Scales with add-ons | - | Large knowledge bases |
Kendra pricing is per-unit per-hour (billed as monthly rate), not per query beyond the included quota. This makes it cost-predictable but expensive relative to query volume for light usage. If your use case is fewer than 5,000 queries/day, evaluate whether Bedrock Knowledge Bases + OpenSearch Serverless might be cheaper.
Interview Focus Points
- 1What is the difference between Amazon Kendra and Amazon OpenSearch Service? When would you use each?
- 2How does Kendra handle document-level access control (ACL) for secure enterprise search?
- 3What types of answers does Kendra return and how do you configure FAQs vs document search?
- 4How does Kendra compare to Bedrock Knowledge Bases for a RAG use case?
- 5How would you keep a Kendra index synchronized with a SharePoint intranet that gets updated daily?
- 6What is the cost structure of Kendra and when does it become expensive?