Bedrock Chat: Open Source RAG Bots with a Bot Store and Multi-Tenant Knowledge Bases

If you've been playing with generative AI on AWS, you've probably seen the Bedrock service. It's powerful, but building a proper RAG (Retrieval Augmented Generation) system on top of it — complete with user management, knowledge bases, and a way to share your bots — is a lot of plumbing.

AWS just open sourced their internal reference implementation for this exact use case. It's called Bedrock Chat, and it's on GitHub. No frills, no sales pitch — just code that solves a real problem.

What It Does

Bedrock Chat is a full-stack application built on AWS Bedrock that lets you:

• Create and manage chatbot instances that use RAG to answer questions from your own data
• Build a "bot store" where users can discover, install, and use bots created by other people
• Manage multi-tenant knowledge bases — meaning different users or teams can have their own isolated content sources
• Use any of the foundation models available in Bedrock (Claude, Llama, Mistral, etc.)

Think of it as an internal AI assistant platform for your organization, or a foundation for building your own chatbot marketplace.

Why It’s Cool

A few things stand out here that make this more than just another RAG demo.

The bot store pattern is practical. Most RAG projects stop at "you can ask questions about your documents." Bedrock Chat adds a layer where users can publish their bots, browse others, and install what they need. This directly maps to how teams actually work — someone in legal builds a contracts bot, someone in support builds a FAQ bot, and everyone else just installs them.

Multi-tenant knowledge bases are handled properly. Each bot can have its own knowledge base, and users can only access what they're supposed to. The repo includes a well-structured setup with IAM roles, S3 buckets, and Aurora PostgreSQL for storing both chat history and vector data. No data leaking between tenants.

It's actually deployable. The repo includes a CDK (Cloud Development Kit) stack that provisions everything. One command and you have a working chat application with authentication (Cognito), API (API Gateway + Lambda), and the frontend (React). You're not stitching together five different tutorials.

Flexible model choice. Because it's on Bedrock, you can swap between models without changing the application code. Claude for long-form answers, Llama for lower latency, or even fine-tuned models if you have them.

How to Try It

The project lives at github.com/aws-samples/bedrock-chat. It's an AWS Sample repo, so expect clean code but also the usual CDK + TypeScript stack.

To get started:

git clone https://github.com/aws-samples/bedrock-chat.git
cd bedrock-chat

Make sure you have the AWS CLI configured and the CDK CLI installed. Then:

npm ci
cdk bootstrap
cdk deploy

This will provision everything in your AWS account. The README has a full deployment guide, including prerequisites like enabling Bedrock access for specific models.

Once deployed, you'll get a URL for the web UI. From there you can create your first bot, upload documents, and start chatting. The bot store shows up automatically as a separate page in the UI.

Final Thoughts

This is one of those repos where you look at it and think "I could have built that, but now I don't have to." It's not revolutionary — it's a solid, well-architected implementation of a pattern that a lot of teams need. If you're already on AWS and trying to figure out how to give your org internal AI assistants without building everything from scratch, this is worth cloning today.

The multi-tenant setup alone saves weeks of security and infrastructure work. And the bot store pattern is smart enough that you could extend it to support approval workflows, ratings, or even monetization if you're building something for external users.

Clone it, deploy it, and let your team start playing. That's what open source is for.

Found this useful? Follow @githubprojects for more developer tools and open source projects.