🚀 Add AWS Bedrock RAG pipeline for OpenWebUI

[weisser-dev]

Summary
This PR introduces a new manifold pipeline that integrates AWS Bedrock Knowledge Bases into OpenWebUI. It allows users to select from multiple knowledge bases and perform Retrieval-Augmented Generation (RAG) using Amazon Bedrock and large language models such as Claude 3.5 Sonnet.

Features

• ✅ Manifold pipeline supporting multiple AWS knowledge bases
• 🔑 Configurable via environment variables:
• AWS_BEDROCK_KB_IDS and AWS_BEDROCK_KB_NAMES (semicolon-separated)
• AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION
• Optional: BEDROCK_RUNTIME_ENDPOINT_URL and BEDROCK_AGENT_RUNTIME_ENDPOINT_URL (for custom VPC access)
• 📚 Uses bedrock-agent-runtime.retrieve() for vector search
• 🧠 Sends context + user question to Claude via bedrock-runtime.invoke_model
• 🧩 Fully integrated with the OpenWebUI valves config (editable per pipeline)

Example Use Case
This pipeline enables RAG workflows similar to those on OpenAI or Azure OpenAI, but using Amazon’s native LLM infrastructure and vector store. It’s particularly useful for:

• Organizations using AWS Bedrock Knowledge Bases as their primary RAG backend
• Private cloud/VPC setups with custom endpoints
• Teams requiring dynamic pipeline selection in OpenWebUI

Notes

• No streaming support yet – output is returned as a single final response.
• Claude model is used by default (anthropic.claude-3-5-sonnet-20240620-v1:0), but can be changed via valves.

[weisser-dev]

@tjbck I think this could be a nice feature for some of your users :) with that you have the option to directly add aws rag systems to openwebui and also multiple systems