Strands Agents Deep Agent Research

A production-ready deep research agent powered by Strands DeepAgents, deployed on AWS Bedrock AgentCore using Terraform.

Overview

This project provides a multi-agent research system that:

• Conducts comprehensive research on any topic using parallel subagents
• Performs intelligent web searches via Linkup/Tavily APIs
• Synthesizes findings with automated citation management
• Deploys as a serverless agent on AWS Bedrock AgentCore
• Stores research outputs in S3

Architecture

Langfuse traces Example

Project Structure

strands-agents-deepresearch/
├── deepresearch/               # Agent source code
│   ├── deepresearch/           # Main package
│   │   ├── main.py             # Agent entry point
│   │   ├── config.py           # Configuration
│   │   ├── prompts/            # Agent prompts
│   │   ├── tools/              # Custom tools
│   │   └── utils/              # Utilities
│   ├── runtime.py              # AgentCore runtime handler
│   ├── pyproject.toml          # Python dependencies
│   └── README.md               # Detailed agent documentation
├── terraform/                  # Infrastructure as Code
│   ├── main.tf                 # Root module
│   ├── variables.tf            # Input variables
│   ├── outputs.tf              # Output values
│   ├── terraform.tfvars.example
│   └── modules/
│       └── agentcore/          # AgentCore deployment module
├── invoke_runtime.py           # CLI to invoke deployed agent
└── README.md                   # This file

Prerequisites

• Python 3.13+
• UV package manager
• AWS CLI configured with appropriate credentials
• Terraform 1.0+
• API keys for Linkup and/or Tavily stored in AWS Secrets Manager

Quick Start

1. Clone and Setup

cd strands-agents-deepresearch

# Create and activate virtual environment
uv venv
source .venv/bin/activate

# Install dependencies
cd deepresearch
uv sync

2. Configure Secrets in AWS

Store your API keys in AWS Secrets Manager:

aws secretsmanager create-secret --name linkup/api-key --secret-string "your-linkup-key"
aws secretsmanager create-secret --name tavily/api-key --secret-string "your-tavily-key"

3. Deploy with Terraform

cd terraform

# Create your configuration
cp terraform.tfvars.example terraform.tfvars
# Edit terraform.tfvars with your values

# Deploy
terraform init
terraform plan
terraform apply

4. Invoke the Agent

# From project root
python invoke_runtime.py \
  --agent-arn "arn:aws:bedrock-agentcore:REGION:ACCOUNT:runtime/AGENT_ID" \
  --prompt "Research the current state of quantum computing in 2025" \
  --region us-east-1

Terraform Configuration

Required Variables

Variable	Description
agent_name	Name for the agent (used in resource naming)
region	AWS region for deployment

Optional Variables

Variable	Default	Description
agent_description	"DeepSearch Agent..."	Agent description
bucket_name	Auto-generated	S3 bucket for deployment package
role_arn	Default AgentCore role	IAM role ARN
network_mode	"PUBLIC"	Network access mode
enable_memory	true	Enable AgentCore memory
create_outputs_bucket	true	Create S3 bucket for outputs
outputs_retention_days	90	S3 output retention period
idle_runtime_session_timeout	60	Session idle timeout (seconds)
max_lifetime	1000	Max runtime lifetime (seconds)

Example terraform.tfvars

agent_name  = "deepsearch-prod"
region      = "us-east-1"

tags = {
  Environment = "production"
  Project     = "deepsearch"
}

Local Development

For local development and testing without AgentCore:

cd deepresearch
source ../.venv/bin/activate

# Set environment variables
export LINKUP_API_KEY="your-key"
export BYPASS_TOOL_CONSENT="true"

# Run directly
python -m deepresearch.main --prompt "Your research query"

Invoking the Deployed Agent

The invoke_runtime.py script provides a CLI to invoke the deployed agent:

python invoke_runtime.py \
  --agent-arn "arn:aws:bedrock-agentcore:us-east-1:123456789:runtime/deepsearch-prod" \
  --prompt "Compare the environmental impact of electric vs hydrogen vehicles" \
  --region us-east-1 \
  --session-id "my-session-123"  # Optional: for conversation continuity

Agent Capabilities

The DeepSearch agent excels at:

• Multi-perspective research: Analyzing topics from multiple viewpoints
• Comparative analysis: Comparing technologies, approaches, or concepts
• Current events: Researching recent developments and news
• Technical deep-dives: In-depth exploration of technical topics

See deepresearch/README.md for detailed agent documentation.

Outputs

Research outputs are stored in S3 with the following structure:

s3://outputs-bucket/
├── {session_id}/
│   ├── final_report.md          # Synthesized research report
│   ├── research_findings/       # Individual subagent findings
│   └── sources/                 # Source documents with URLs

Monitoring

• Logs: Available in CloudWatch Logs
• Telemetry: OpenTelemetry traces via Langfuse integration
• Metrics: AgentCore runtime metrics in CloudWatch

Cost Considerations

Typical costs per research query:

• Claude API: $0.50-2.00 depending on complexity
• AgentCore Runtime: Pay-per-use serverless pricing
• Search APIs: Based on your Linkup/Tavily plan
• S3 Storage: Minimal for text outputs

License

See LICENSE file for details.