agent-builder

Knowledge Base Integration

Strands Agent Builder leverages Amazon Bedrock Knowledge Bases to store and retrieve custom tools, agent configurations, and development history.

Set up your Knowledge Base

Prerequisites

• AWS account with IAM user
• Access to Amazon Bedrock console
• Permissions to create IAM roles and S3 buckets

Console Setup (Recommended)

1. Access Amazon Bedrock Console

   • Sign in to AWS Management Console
   • Navigate to Knowledge bases in the left panel

2. Create Knowledge Base

   • Click Create → Knowledge base with vector store
   • Enter a name and description for your knowledge base

3. Configure IAM Role

   • Choose Create and use a new service role (recommended)
   • Or select an existing role with Bedrock permissions

4. Set Up Data Source

   • Choose your data source type, select Custom for Strands Agent
   • Configure connection details

5. Configure Embeddings

   • Select an embeddings model (e.g., Amazon Titan Text Embeddings V2)
   • Choose embedding type: float32 (precise) or binary (cost-effective)

6. Choose Vector Store

   • Quick create (recommended): Let Bedrock create and manage the vector store
   • Custom: Use existing OpenSearch Serverless, Aurora PostgreSQL, or S3 Vectors

7. Review and Create

   • Review all configurations
   • Click Create knowledge base
   • Wait for creation to complete (status: "Ready" or "Available")

8. Sync Data Source

   • Select your knowledge base
   • Click Sync in the data source section
   • Wait for initial sync to complete

API Setup (Advanced)

Create via AWS CLI or SDK:

# Example using AWS CLI
aws bedrock-agent create-knowledge-base \
  --name "MyKnowledgeBase" \
  --description "Strands Agent Builder KB" \
  --role-arn "arn:aws:iam::ACCOUNT:role/AmazonBedrockExecutionRoleForKnowledgeBase" \
  --knowledge-base-configuration '{
    "type": "VECTOR",
    "vectorKnowledgeBaseConfiguration": {
      "embeddingModelArn": "arn:aws:bedrock:us-east-1::foundation-model/amazon.titan-embed-text-v2:0"
    }
  }' \
  --storage-configuration '{
    "type": "OPENSEARCH_SERVERLESS",
    "opensearchServerlessConfiguration": {
      "collectionArn": "arn:aws:aoss:us-east-1:ACCOUNT:collection/YOUR_COLLECTION",
      "vectorIndexName": "bedrock-knowledge-base-index",
      "fieldMapping": {
        "vectorField": "bedrock-knowledge-base-default-vector",
        "textField": "AMAZON_BEDROCK_TEXT_CHUNK",
        "metadataField": "AMAZON_BEDROCK_METADATA"
      }
    }
  }'

For more set up details, please see AWS Bedrock Knowledgebase

Get Your Knowledge Base ID

After creation, find your Knowledge Base ID:

• Console: Copy from the knowledge base details page
• CLI: aws bedrock-agent list-knowledge-bases

The ID format: ABCDEFGHIJ (10 characters)

Use your KnowledgeBase

# Load and extend tools from your knowledge base
strands --kb YOUR_KB_ID "Load my data_visualizer tool and add 3D plotting capabilities"

# Or set a default knowledge base via environment variable
export STRANDS_KNOWLEDGE_BASE_ID="YOUR_KB_ID"
strands "Find my most recent agent configuration and make it more efficient"

Features:

• 🔄 Retrieve previously created tools and agent configurations
• 💾 Persistent storage for your custom tools and agents
• 🛠️ Ability to iteratively improve tools across sessions
• 🔍 Find and extend tools built in previous sessions

Model Configuration

Optimized Defaults

Strands comes with optimized, maxed-out configuration settings for the Bedrock model provider:

{
    "model_id": "us.anthropic.claude-sonnet-4-20250514-v1:0",
    "max_tokens": 32767,
    "boto_client_config": {
        "read_timeout": 900,
        "connect_timeout": 900,
        "retries": {
            "max_attempts": 3,
            "mode": "adaptive"
        }
    },
    "additional_request_fields": {
        "anthropic_beta": ["interleaved-thinking-2025-05-14"],
        "thinking": {
            "type": "enabled",
            "budget_tokens": 2048
        }
    }
}

These settings provide:

• Claude Sonnet 4 (latest high-performance model)
• Maximum token output (32,768 tokens)
• Extended timeouts (15 minutes) for complex operations
• Automatic retries with adaptive backoff
• Interleaved thinking capability for real-time reasoning during responses
• Enabled thinking capability with 2,048 token budget for recursive reasoning

You can customize these values using environment variables:

# Maximum tokens for responses
export STRANDS_MAX_TOKENS=32000

# Budget for agent thinking/reasoning
export STRANDS_BUDGET_TOKENS=1024

Custom Model Provider

You can configure strands to use a different model provider with specific settings by passing in the following arguments:

strands --model-provider <NAME> --model-config <JSON|FILE>

As an example, if you wanted to use the packaged Ollama provider with a specific model id, you would run:

strands --model-provider ollama --model-config '{"model_id": "<ID>"}'

Strands Agent Builder is packaged with bedrock and ollama.

If you have implemented a custom model provider (instructions) and would like to use it with strands, create a python module under the directory "$CWD/.models" and expose an instance function that returns an instance of your provider. As an example, assume you have:

$ cat ./.models/custom_model.py
from mymodels import CustomModel

def instance(**config):
    return CustomModel(**config)

You can then use it with strands by running:

$ strands --model-provider custom_model --model-config <JSON|FILE>

Custom System Prompts

# Via environment variable
export STRANDS_SYSTEM_PROMPT="You are a Python expert."

# Or local file
echo "You are a security expert." > .prompt

🌍 Environment Variables Configuration

Strands Agent Builder also provides customization through environment variables:

TrulsElg /

Python-AI-Agent

Toy example of an agentic AI code assistant