Key takeaway
- Amazon Kendra is in maintenance mode and closed to new customers. See the benefits existing customers keep, the costs and risks, and when to move.
What are the benefits of Amazon Kendra for existing customers?
Amazon Kendra still gives existing customers natural language search over enterprise content, connectors to common repositories and access filtering by user or group. Since 30 June 2026 it has been in maintenance mode, so no new features are planned, but AWS says the service remains fully supported with bug fixes and security updates for existing customers (AWS Kendra availability change).
It is also closed to new customers: the AWS page gives 30 July 2026 for that, and AWS recommends Amazon Bedrock Managed Knowledge Base for new search applications. The practical question for a CTO or platform lead is therefore which benefits are worth keeping, and when to move. For setup steps and cost see our Amazon Kendra Implementation Guide.
What has changed for Amazon Kendra in 2026?
| Item | Status | Source |
|---|---|---|
| Maintenance mode | Effective 30 June 2026, no new feature development | AWS |
| New customers | No longer open; the detailed text gives 30 July 2026 (the page banner is undated) | AWS |
| Existing customers | Fully supported, with bug fixes and security updates | AWS |
| Recommended alternative | Amazon Bedrock Managed Knowledge Base | AWS |
| End-of-support date | None stated on the AWS page | AWS page read 12 Nov 2026 |
We have not found an end-of-support date, so treat the timing of any move as your own decision rather than an AWS deadline.
Which Kendra benefits still matter to an existing estate?
| Benefit | What it gives you | Why it matters now |
|---|---|---|
| Natural language questions | Handles factoid, descriptive and keyword or natural language queries (per AWS) | Users keep asking in their own words; check the answers against your own test set |
| Repository connectors | Kendra has 32+ native connectors; the managed knowledge base has 7 (per AWS) | Connector coverage is the most likely blocker for a move |
| Access filtering | Results can be filtered by user or group; you authenticate and authorise users | Keep the same permission model if you migrate |
| GenAI index | Designed for retrieval augmented generation; usable with Amazon Q Business and Bedrock knowledge bases (per AWS) | Lets existing content feed generative applications |
| Features with no direct equivalent | Query suggestions, faceted search, custom synonyms, spell checking, incremental learning, custom document enrichment | AWS lists workarounds that you must build yourself |
The query types, index editions and access filtering are described on the AWS Kendra overview. Semantic matching is the idea behind these benefits; our post on the semantic search engine explains it in more depth.
How should you run an existing Kendra index in 2026?
What does it cost to keep it?
AWS states that after the free trial you are charged for all provisioned indexes, even empty ones with no queries, and that syncing documents from data sources adds charges. Retire unused indexes and Developer edition test indexes first. The Developer edition is described by AWS as for proof of concept work, not production.
What about governance and security?
You remain responsible for authenticating and authorising users, so keep the access rules documented and test that restricted documents stay hidden. Maintenance mode brings security updates but no new controls.
How do enterprise agents use Kendra today?
Kendra can act as the retrieval layer for agents. Our post on AI agents with Amazon Kendra shows the pattern. If you are choosing retrieval for a new agent, AgentSearch surfaces relevant unstructured company knowledge for agents and staff, and ElixirData keeps governed context, versioned policy and decision traces in one decision layer above AI agents. ElixirClaw is an agentic OS platform to build, orchestrate and govern AI agents and automation workflows while keeping humans in control.
When should you move from Kendra to a Bedrock knowledge base?
| Situation | Suggested action |
|---|---|
| Stable internal search, supported connectors, few custom features | Stay, review yearly, retire idle indexes |
| Relies on connectors outside the managed knowledge base list | Export content to S3 and test a sync pipeline, or stay until you have a plan |
| Depends on facets, synonyms, query suggestions or enrichment | Budget for the workarounds AWS lists before moving |
| Planning a new search or RAG application | Start on the AWS recommended path, not Kendra (closed to new customers) |
Before moving, AWS suggests a parallel test comparing ranking quality (NDCG and MRR), latency, throughput and completeness. The managed knowledge base always uses hybrid (keyword plus semantic) search and has no semantic chunking; AWS suggests fixed-size chunks of 200 tokens with 30% overlap as a starting point. It supports 7 connectors against Kendra's 32+.
When is staying on Kendra the wrong choice?
Staying is a poor fit if you need capabilities AWS will not build, or if an index sits idle and costs money. AWS states that no new features are being developed, so there is no feature roadmap to wait for.
To map one search or retrieval decision your agents should make, and the context and policy it depends on, book a Context Workshop. See what happens in the workshop.
Amazon Kendra FAQs
Is Amazon Kendra still available?
Existing customers can keep using it. It entered maintenance mode on 30 June 2026 and is closed to new customers, with 30 July 2026 given in the AWS text.
Will AWS keep fixing Kendra?
AWS says the service stays fully supported with bug fixes and security updates for existing customers, but new feature requests are no longer considered.
What does AWS recommend instead of Kendra?
Amazon Bedrock Managed Knowledge Base, for migrating existing applications and for new search applications.
Is there an end-of-support date for Kendra?
The AWS page read on 12 November 2026 does not state one.
What is the biggest migration risk?
Connector coverage and missing features: Kendra has 32+ connectors against 7, and features such as facets, synonyms and query suggestions need workarounds.
Do empty Kendra indexes cost money?
Yes. AWS states that after the free trial you are charged for all provisioned indexes, even empty ones.
Follow XenonStack on LinkedIn: XenonStack on LinkedIn for new guides on agentic AI, data and platform engineering.
Explore the XenonStack platform family: ElixirData · ElixirClaw · ElixirHub · NexaStack · MetaSecure · Neural AI · AgentGRC · AgentSRE · AgentAnalyst · AgentInstruct · AgentRAI · AgentScrum · AgentSearch · AgentForce · AI Aviator · SAIF Aviator · CohortIQ · XenonStack AI · XenonStack UAE
