How to Intermediate · 3 min read

How to create knowledge base in AWS Bedrock

Quick answer

To create a knowledge base in AWS Bedrock, use the boto3 bedrock-runtime client to upload and index your documents, then query them with the converse API. This involves preparing your documents, storing them in a vector store or database, and invoking Bedrock models for retrieval-augmented generation.

PREREQUISITES

Python 3.8+
AWS account with Bedrock access
AWS CLI configured or environment variables for AWS credentials
pip install boto3

Setup

Install the boto3 library and configure AWS credentials with Bedrock access. You can set credentials via ~/.aws/credentials or environment variables AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

bash

pip install boto3

Step by step

This example shows how to create a simple knowledge base by uploading documents, embedding them externally (e.g., using OpenAI embeddings), storing vectors in a database, and querying with AWS Bedrock's converse API for retrieval-augmented generation.

python

import os
import boto3
import json

# Initialize Bedrock runtime client
client = boto3.client('bedrock-runtime', region_name='us-east-1')

# Example: documents to add to knowledge base
documents = [
    {"id": "doc1", "text": "AWS Bedrock enables easy access to foundation models."},
    {"id": "doc2", "text": "You can build knowledge bases by combining embeddings and Bedrock models."}
]

# Normally, you would embed these documents with an embedding model and store vectors in a vector DB
# For demo, we simulate a simple retrieval by matching keywords

# User query
query = "How to create a knowledge base in AWS Bedrock?"

# Compose prompt with retrieved context (simulate retrieval)
retrieved_context = "AWS Bedrock enables easy access to foundation models."

prompt = f"Use the following context to answer the question.\nContext: {retrieved_context}\nQuestion: {query}\nAnswer:"

# Call Bedrock converse API with Anthropic Claude model
response = client.converse(
    modelId="anthropic.claude-3-5-haiku-20241022",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": prompt}
    ]
)

answer = response['output']['message']['content'][0]['text']
print("Answer:", answer.strip())

output

Answer: To create a knowledge base in AWS Bedrock, you prepare your documents, embed them for retrieval, and then use Bedrock models like Claude to answer queries based on the retrieved context.

Common variations

Use different Bedrock models by changing modelId, e.g., amazon.titan-text-express-v1.
Integrate external vector databases like Pinecone or FAISS for scalable retrieval.
Use async calls with aiobotocore for concurrency.

Troubleshooting

If you get AccessDeniedException, verify your AWS Bedrock permissions and region.
For ModelNotFoundException, confirm the modelId is correct and available in your region.
Timeouts may require increasing client timeout or retry logic.

✅

Key Takeaways

Use boto3 bedrock-runtime client to interact with AWS Bedrock models.
Combine document embeddings and vector search with Bedrock's converse API for knowledge base queries.
Ensure AWS credentials and Bedrock permissions are correctly configured for API access.

Verified 2026-04 · anthropic.claude-3-5-haiku-20241022, amazon.titan-text-express-v1

Verify ↗