Running an event stand with Strands Agents? Here is your demo kit
A ready-to-deploy event stand kit for Strands Agents: 11 live demos on AWS, split-screen UI showing the agent loop in real time. Clone it, deploy with your AWS credits, fork it for your own demos.
You love Strands Agents and you have a stand at a AWS Community Day, a meetup, or any tech event. You want people to stop, interact, and walk away with a clear picture of what an AI agent actually does. Not a slide. Not a diagram. Something they can type into themselves.
This post introduces a sample repo built exactly for that: a split-screen web app you deploy once and leave running at your stand. Visitors type a message, the left side shows the agent answer, and the right side shows every reasoning step, tool call, blocked action, and token spent in real time. The agent loop becomes visible to anyone who walks up, no prior knowledge required.
The Strands Agents framework is open source. This repo is a public sample that runs on AWS using Amazon Nova Pro, so you can run it perfectly with AWS console credits. When the event is over,
cdk destroy removes everything cleanly, with no resources left behindWhat visitors see
The app is a split-screen web interface. On the left: a chat. On the right: a live "under the hood" panel that shows the agent loop, every tool call, blocked actions, multi-agent handoffs, and token costs, updating as the agent runs.
When a visitor sends a message, they watch the agent reason through the problem, decide which tool to call, execute it, and produce an answer, cycle by cycle. When you run the hooks demo, they watch the model try to call a forbidden tool and a security hook block it with a red card before the agent can even respond.
The UI ships in English, Spanish, and Portuguese. If your community speaks a different language, the localization lives. Add your language and submit a pull request.
11 demos, each showing one agent concept
The kit ships with 11 independent demos. You do not need to run all of them; pick two or three that fit your stand.
| # | Demo | What it shows |
|---|---|---|
| 01 | Live Agent Loop | The Reason, Tool, Respond cycle animating in real time |
| 02 | Structured Output | Free text in, validated typed record out (Pydantic) |
| 03 | Hooks: the Guardian | Visitors try to talk the agent into dangerous actions; hooks block them live |
| 04 | Human-in-the-loop | The agent searches flights alone but freezes before booking; you approve or reject |
| 05 | Agent X-Ray | Full observability: business attributes, per-tool latency, ground-truth ledger |
| 06 | Live Swarm | Researcher, analyst, and writer collaborating with animated handoffs |
| 07 | Stop Wasting Tokens | Same question, two agents: ~27k tokens vs ~1.4k (~95% reduction) on live meters |
| 08 | Graph Pipeline | A deterministic DAG: brainstormer, fact-checker and critic, editor |
| 09 | Agents as Tools | A concierge delegates to specialist agents called like functions |
| 10 | Memory Poisoning | Plant a malicious note in turn 1, trigger exfiltration in turn 2; tool boundary blocks it |
| 11 | Chaos Testing | Same question twice: without the harness the agent reports 999 °C; with it, correct |
| 🤖 | Robots | soon |
Each demo is a self-contained Python agent in its own directory, deployed on its own Amazon Bedrock AgentCore Runtime . If one demo breaks, the others keep working.
Deploy it
1
2
3
git clone https://github.com/elizabethfuentes12/strands-agents-demo-sample-for-aws
cd strands-agents-demo-sample-for-aws
KIOSK_PASSWORD='<choose-a-strong-password>' ./deploy.shdeploy.sh runs the full pipeline: Python virtualenv, CDK bootstrap check, 12 CloudFormation stacks (shared base plus one per demo), Cognito kiosk user, web configuration, and a smoke test. If the smoke test prints SMOKE TEST OK, the full pipeline is live.Set your own
KIOSK_PASSWORD. The demo runs on a public CloudFront URL during the event, so use a password you chose and can share with your team on the day.Prerequisites: Python 3.11+, Node 20+, AWS CLI v2, AWS CDK v2 (via
npx cdk), uv , and Amazon Bedrock model access to Amazon Nova Pro (us.amazon.nova-pro-v1:0) in your account.After deploy you get a CloudFront URL and the kiosk credentials. Open it, log in, and you are ready.
Use it as your starting point
The demos are not read-only examples. Fork the repo and adapt it:
- Swap the model. Every demo reads
MODEL_IDfrom an environment variable. One CDK config change switches all 11. - Add your own demo. Each demo is one Python file plus a
requirements.txt. Adding a new one is a one-line change toDEMOSininfra/app.py, no new CDK construct needed. - Change the system prompts. Every
SYSTEM_PROMPTlives at the top of eachagent.py, clearly labeled. Adjust the persona, domain, or tools to match your use case. - Add your language. Add an entry to
web/src/i18n.ts, extend theLangtype, and submit a PR.
Cost
At 1,000 interactions, the stack costs roughly $5, dominated by model tokens (~$3.77) with the rest in AgentCore compute and AppSync operations. When the event ends,
cdk destroy removes everything with no orphaned resources and no ongoing cost.Full cost breakdown in the README .
Try it now
If you build something new with it, a demo for your use case, a localization for your community, an adaptation for a different domain, contributions are welcome .
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article