Bedrock AgentCore Chatbot

FAQ chatbot powered by LangChain + LangGraph, deployed on Amazon Bedrock AgentCore Runtime, with a Streamlit frontend and CloudWatch tracing.

Demo

Streamlit UI

CloudWatch Trace View

Project Highlights

• FAQ retrieval with vector search (FAISS + sentence-transformers embeddings)
• Tool-using agent behavior for structured Q&A responses
• AgentCore memory-enabled runtime integration
• Streamlit chat app that invokes the deployed runtime (not local direct invoke)
• CloudWatch-compatible runtime invocation path for request tracing

Project Structure

.
|-- streamlit_app.py          # Streamlit frontend (invokes AgentCore runtime via boto3)
|-- agentcore_memory.py       # Memory-enabled AgentCore runtime entrypoint
|-- agentcore_runtime.py      # AgentCore runtime variant
|-- langgraph_agent.py        # Local LangGraph/agent variant
|-- lauki_qna.csv             # FAQ knowledge base
|-- .bedrock_agentcore.yaml   # AgentCore deployment/runtime metadata
|-- pyproject.toml            # Project dependencies
|-- assets/
|   |-- frontend_ui.png
|   `-- cloudwatch_trace.png

Prerequisites

• Python >=3.13
• AWS credentials configured for the target account/region
• Access to Bedrock AgentCore runtime and memory resources
• GROQ_API_KEY and any required model/provider credentials in .env

Quick Start

1. Install dependencies

uv sync

2. Ensure environment variables are set (example)

$env:AWS_REGION="us-west-2"
$env:AWS_DEFAULT_REGION="us-west-2"
$env:AGENT_RUNTIME_ARN="arn:aws:bedrock-agentcore:...:runtime/your-runtime-id"

3. Run the frontend

uv run streamlit run streamlit_app.py

4. Open the app in your browser (typically http://localhost:8501)

How It Works

1. User asks a question in Streamlit.
2. streamlit_app.py resolves AgentCore runtime ARN and region.
3. App calls bedrock-agentcore:InvokeAgentRuntime with:
   • prompt
   • actor_id
   • thread_id
4. Deployed runtime executes agentcore_memory.py agent flow.
5. Response is returned to UI and trace data is available in CloudWatch.

Notes on Memory and Tracing

• actor_id and thread_id are captured in the UI sidebar for session continuity.
• Runtime calls are sent through AWS AgentCore API, enabling CloudWatch trace visibility.
• If traces do not appear, verify:
  • correct runtime ARN
  • runtime observability enabled in AgentCore config
  • IAM permissions for invoke and logging
  • matching AWS account and region

Security Reminder

• Do not commit real secrets in .env (API keys/tokens).
• Rotate exposed keys and prefer AWS Secrets Manager or environment injection in deployment.