Build multi-tenant agentic chat applications on enterprise data with Amazon Bedrock Managed Knowledge Base
aws ·
The article presents an architecture for building multi-tenant agentic chat applications that let users upload documents and ask questions, using Amazon Bedrock Managed Knowledge Base. It emphasizes enforcing tenant isolation through verified user identities from Amazon Cognito JWTs rather than client-supplied values, and applying metadata filters on every retrieval hop. The solution includes components such as API Gateway, Lambda, SQS, DynamoDB, S3, and CloudFront, with two main data flows: document ingestion and conversational retrieval. For ingestion, documents are sent directly via a custom connector, either inline or through S3 for larger files, and tagged with a user_id. The indexing lifecycle runs through states from STARTING to INDEXED, with documents becoming text-queryable at TEXT_INDEXED and fully indexed later, especially for PDFs. For retrieval, the application calls the agentic retrieval API, which plans multi-hop queries and returns cited, grounded answers. The authors recommend a shared knowledge base with metadata filtering for many end users to avoid per-tenant infrastructure costs, and they provide a reference repository for deployment.
aws