What’s new here

This repo provides two concrete compute paths: full Kubernetes on EKS with Helm charts, or a lighter ECS-plus-AgentCore Runtime setup. Both share the same agent code, the same FastAPI interface on port 8080, and the same LiteLLM-based LLM gateway, which runs in the cluster on EKS and in ECS on the other path. Swapping compute layers doesn’t require rewriting agents.

What it does

The repo is a monorepo organized into layered infrastructure stacks. Five Terraform stacks under infrastructure/ build a VPC foundation, then either an EKS cluster or an ECS/AgentCore environment on top, plus a knowledge layer for RAG. Helm charts in k8s/ deploy agents and platform services into EKS. Application code in src/agentic_platform/ holds five sample agents (built with Strands or LangGraph), a shared core library, MCP servers, and gateway services for LLM calls, memory, and retrieval. Auth runs through JWT tokens validated against an IDP cert; telemetry ships to X-Ray, CloudWatch, and OpenSearch via OpenTelemetry.

The labs/ directory contains five progressive modules with 25+ Jupyter notebooks covering prompt engineering, agentic patterns, memory, MCP, multi-agent systems, and deployment. Modules 1 through 4 run locally; only Module 5 requires the platform to be deployed.

A tree of AGENTS.md files at the root and in each major directory gives coding agents (Claude Code, Cursor, Kiro) a navigable map of the codebase.

Who it’s for

Platform engineers and application teams at AWS shops who need a concrete starting point for running agents in production. It’s most useful if you’re deciding between managing your own EKS cluster versus handing container orchestration to AgentCore, or if you want to see how gateway services, auth, and telemetry fit together before building your own version.

Try it

Prerequisites: Python 3.12, uv, Docker.

# Clone and install
git clone https://github.com/aws-samples/sample-agentic-platform.git
cd sample-agentic-platform
uv sync
 
# Start supporting services (Postgres, Redis, LiteLLM, Memory Gateway)
make dev:deps
 
# Run an agent
make dev agentic_chat
 
# Run an MCP server
make dev:mcp bedrock_kb_mcp_server
 
# Stop supporting services when done
make dev:deps-stop

Run make help to see all available commands. For AWS deployment, see DEPLOYMENT.md. Labs 1-4 only need uv sync then uv run jupyter lab.

How mature is it

141 stars, 58 forks, 9 contributors. Created April 2025, last pushed September 2026. No versioned releases. 30 open issues and pull requests. 1 commit in the last 90 days, although the README calls the sample actively maintained. Licensed MIT-0 (no attribution required).