Deploy a full LLM platform on AWS using only managed services

What it does

Bedrock Chat is a full-stack, multilingual chat interface built entirely on AWS managed services. Users can converse with foundation models, create custom bots grounded in uploaded documents via RAG, share them through an internal bot store, and publish them as standalone APIs. Administrators get usage analytics, API management, and governance controls over who can build and access bots.

The interesting bit

The project’s real value is in the plumbing. It works around the 100-Knowledge-Base-per-account limit by attaching Bot IDs as metadata filters to a shared OpenSearch Serverless index, giving each bot its own tenant slice without provisioning separate vector databases. The backend is a FastAPI application wrapped with the AWS Lambda Web Adapter, running on API Gateway and Lambda.

Key highlights

• Multi-tenant RAG out of the box: new bots share a single OpenSearch Serverless index, sidestepping AWS account limits by filtering on Bot ID metadata.
• Custom bots can be published as standalone APIs or shared via a built-in bot store, gated by Cognito user groups.
• Includes an agent framework for multi-step tasks that can retrieve information from external tools.
• Administrative dashboard for usage analytics, marking essential bots, and managing published APIs.
• Deployment stays entirely inside the AWS network; the architecture uses Bedrock, Lambda, DynamoDB, and CloudFront with no external API calls.

Caveats

• Upgrading from V2 to V3 is breaking; the README warns that existing bots will become unusable without a careful migration.
• The default deployment targets the latest development version, which the README admits can be unstable—if the frontend fails to appear, you are advised to pin an older version.
• Out of the box, the stack allows anyone with the URL to self-register; production use requires manually tightening Cognito, IP ranges, and email domain restrictions.

Verdict

Worth a look if you need an AWS-native, multi-user LLM reference architecture with RAG and governance controls already wired up. Skip it if you are not all-in on AWS managed services or just need a lightweight, single-tenant chatbot.