Production-ready multi-agent framework for building scalable data analysis workflows without infrastructure headaches
- [2026/03] Ops Dashboard - job tracking, email notifications, and admin dashboard with Cognito authentication (details)
- [2026/02] Web UI (v1.1.0) - browser-based interface for data upload, analysis, HITL plan review, and report download (details)
- [2025/12] Claude Code-style Skill System - dynamically discover and load specialized skills (PDF, DOCX, XLSX processing, etc.) in Strands Agents with lazy loading for optimal performance (details)
- [2025/12] Human in the Loop (HITL) - review and steer the analysis plan before execution, giving users control over the analysis direction (details)
- [2025/12] Managed AgentCore deployment - production-ready with Bedrock AgentCore Runtime, Custom Code Interpreter (Fargate), and 100% private VPC
- [2025/12] File-based code execution - significantly reduces NameError/ImportError rates compared to REPL-based approaches
- [2025/12] Output token optimization with shared utils scripts - repeatedly used functions are generated once and imported, reducing redundant code generation
Show older updates
- [2025/11] Added per-agent token tracking with detailed metrics - monitor input/output tokens and cache reads/writes for complete cost visibility and optimization
- [2025/11] Added editable DOCX report generation - all analysis results are exportable to fully editable Word documents for easy customization and sharing
- [2025/10] Released Deep Insight Workshop (Korean | English)
- [2025/10] Added support for Claude Sonnet 4.5 with extended thinking and enhanced reasoning capabilities
- [2025/09] Released Deep Insight framework built on Strands SDK and Amazon Bedrock with hierarchical multi-agent architecture
Deep Insight provides a proven hierarchical architecture with Coordinator, Planner, Supervisor, and specialized tool agents. Start with a working production-grade system and customize from there—no need to design from scratch.
Get production-grade multi-agent workflows out of the box with prompt caching, streaming responses, token tracking, and battle-tested performance patterns. Deploy with confidence using architecture validated in real-world scenarios.
Deploy Deep Insight in your own AWS VPC for complete data isolation and control. All data processing happens within your secure VPC environment, with Amazon Bedrock API calls staying in AWS infrastructure—never exposed to the public internet.
Transform complex data analysis into automated insights using hierarchical multi-agent systems built on Strands SDK and Amazon Bedrock.
- 🤖 Advanced Multi-Agent Architecture - Hierarchical workflow with Coordinator, Planner, Supervisor, and specialized tool agents
- 🎨 Full Customization & Control - Modify agents, prompts, and workflows with complete code access
- 🧠 Flexible Model Selection - Choose different Claude models for each agent (Sonnet 4, Haiku 4, etc.) via simple .env configuration
- 💻 Custom Code Interpreter - Flexible code execution from local Python to Fargate-based containers with your own Docker image
- 📊 Transparency & Verifiability - Reports with calculation methods, sources, and reasoning processes
- 🔒 Enterprise-Grade Security - From local development to 100% private VPC with Bedrock AgentCore Runtime
- ⚡ Production Scalability - Concurrent processing with AgentCore MicroVM and auto-scaling Fargate containers
- 🚀 Beyond Reporting - Extend to any agent use case: shopping, support, log analysis, and more
Deep Insight provides three deployment options to match your needs.
Run agents locally or in your VPC with full control:
- ✅ Complete code access to agents, prompts, and workflows
- ✅ Rapid iteration during development (no rebuild required)
- ✅ Simple setup in ~10 minutes
Get Started: ./self-hosted/ | 📖 Self-Hosted README
Production deployment using AWS Bedrock AgentCore Runtime with VPC Private Mode:
- ✅ Bedrock AgentCore Runtime hosting Strands Agent
- ✅ Custom Code Interpreter (ECR + ALB + Fargate)
- ✅ 100% private network (VPC endpoints, no public internet)
Get Started: ./managed-agentcore/ | 📖 Managed AgentCore README
Browser-based interface for non-technical users:
- ✅ Upload data, review plans, download reports from the browser
- ✅ Korean/English language support
- ✅ Internet-facing ALB with VPN CIDR restriction
- ✅ Optional: Ops dashboard for job tracking and email notifications
Get Started: ./deep-insight-web/
| Self-Hosted | Managed AgentCore | Web UI | |
|---|---|---|---|
| Setup Time | ~10 minutes | ~45 minutes | ~15 minutes (after Managed) |
| Agent Hosting | Local/EC2 | Bedrock AgentCore Runtime | Same as Managed |
| Code Execution | Local Python | Custom Code Interpreter (Fargate) | Same as Managed |
| Network | Your choice | 100% Private VPC | VPN-restricted ALB |
| Best For | Development, Testing | Production, Enterprise | Non-technical Users |
📖 Detailed comparison → Security, cost, features, and when to choose each option
Task: "Create a sales performance report for Moon Market. Analyze from sales and marketing perspectives, generate charts and extract insights, then create a docx file. The analysis target is the
./data/Dat-fresh-food-claude.csvfile."Workflow: Input (CSV data file:
Dat-fresh-food-claude.csv) → Process (Natural language prompt: "Analyze sales performance, generate charts, extract insights") → Output (DOCX report with analysis, visualizations, and marketing insights)
📄 English Report | 📄 Korean Report
We welcome contributions! See CONTRIBUTING.md for details.
# Fork the repository on GitHub, then clone your fork
git clone https://github.com/aws-samples/sample-deep-insight.git
cd sample-deep-insight
# Follow the self-hosted setup instructions
cd self-hosted
cd setup/ && ./create-uv-env.sh deep-insight 3.12 && cd ..
# Create feature branch
git checkout -b feature/your-feature-name
# Make changes, test, then commit and push
git add .
git commit -m "Add feature: description"
git push origin feature/your-feature-name
# Open a Pull Request on GitHub- New Agent Types: Add specialized agents for specific domains
- Tool Integration: Create new tools for agent capabilities
- Model Support: Add support for additional LLM providers
- Documentation: Improve guides, examples, and tutorials
- Bug Fixes: Fix issues and improve stability
- Performance: Optimize streaming, caching, and execution
This project is licensed under the MIT License - see the LICENSE file for details.
"Come From Open Source, Back to Open Source"
We believe in the power of open collaboration. Deep Insight takes the excellent work of the LangManus community and extends it with AWS-native capabilities, then contributes those enhancements back to the community.
| Name | Role | Contact |
|---|---|---|
| Gonsoo Moon | AWS Sr. AI/ML Specialist SA | Email · LinkedIn · GitHub · Hugging Face |
| Chloe(Younghwa) Kwak | AWS Sr. Solutions Architect | Email · LinkedIn · GitHub · Hugging Face |
| Yoonseo Kim | AWS Solutions Architect | Email · LinkedIn · GitHub |
| Jiyun Park | AWS Solutions Architect | Email · LinkedIn · GitHub |
| Dongjin Jang, Ph.D. | Previous AWS Sr. AI/ML Specialist SA | Email · LinkedIn · GitHub · Hugging Face |
Built with ❤️ by AWS KOREA SA Team
Empowering enterprises to build customizable agentic AI systems