Amazon Kendra

❓ What is it?

An enterprise search service that understands natural-language questions and returns specific answers (not just links) from content spread across S3, SharePoint, Salesforce, databases, and dozens of other connectors, respecting document ACLs.

💡 Why does it exist?

Keyword search fails when employees ask questions the way humans do ("how many days of parental leave do we get?"). Kendra ranks semantically and extracts the answering passage, cutting time-lost-to-searching.

⏱️ When should you use it?

Use Kendra when the need is SEARCH over enterprise content — intranet search, agent-assist lookups — or as a high-quality retriever feeding RAG (its retrieval API pairs with Bedrock models).

🗺️ Where does it fit?

Above your document repositories: connectors sync content into a Kendra index; applications query it directly, or a generative layer (Bedrock) uses it as the retrieval step and cites the sources.

🔌 How do you integrate it?

Create an index, attach data-source connectors on sync schedules, then call Query for answers/documents or Retrieve for RAG-oriented passages; tune with relevance boosting and FAQs.

🧩 Commonly integrated with

🎯 Exam angle (AIF-C01)

• Kendra = semantic SEARCH with source documents; it retrieves, it does not generate. "Search across company documents in natural language" → Kendra.
• As the retriever in RAG architectures it is an alternative to OpenSearch vector search — with connectors and ACLs as its edge.

📚 Study it in a learning path

More in AI Services