AWS Builder Center

Builder Center

CloudSnack — The Guilt-Free AI Cloud Hygiene Companion People Actually Enjoy Using

CloudSnack — The Guilt-Free AI Cloud Hygiene Companion People Actually Enjoy Using

Discover how an Amazon Bedrock-powered agent turns terrifying AWS bills into café treats and replaces cloud anxiety with guilt-free, 1-click safe remediation recipes. Learn why empathetic design and plain-English error translations make cloud hygiene something developers actually look forward to.

Series: AI (1 article)

  1. 1
    CloudSnack — The Guilt-Free AI Cloud Hygiene Companion People Actually Enjoy Using This article

Hailemichael Tehka Meresa

Cloud/DevOps

e Origin Story: Why Does Cloud Cost Management Feel Like an IRS Audit?

Let’s be brutally honest for a moment: nobody likes opening AWS Cost Explorer on a Monday morning.
You’ve spent an exhilarating weekend building a new microservice prototype, deploying a serverless stack, or experimenting with generative AI models. But by mid-month, an ominous alert email lands in your inbox:
"Billing Alert: Your account 123456789012 has exceeded 85% of your forecasted threshold. Resource arn:aws:ec2:us-east-1:...:volume/vol-03b9841f is incurring charges."
Instantly, the dopamine of building vanishes, replaced by a sinking feeling of dread. You log into the AWS Management Console and face a barrage of 40-row tables, cryptic alphanumeric Resource IDs, and unhelpful cost anomaly graphs. You feel guilty, overwhelmed, and anxious about accidentally breaking a production dependency if you delete the wrong thing.
When Ben Fowler posted this weekend’s challenge with the theme: "Build an agent people actually enjoy using — Small and delightful beats big and clunky," I knew immediately what needed to exist.
We don't need another enterprise corporate dashboard.
We don't need another cold, robotic monitoring bot.
We need CloudSnack: a warm, witty, and deeply empathetic AI cloud hygiene companion powered by Amazon Bedrock that turns terrifying AWS waste into delicious, guilt-free treats and copy-paste safe remediation recipes.

1. What and Who

What CloudSnack Does

CloudSnack is a lightweight, zero-friction conversational AI companion that helps developers maintain squeaky-clean AWS cloud environments without the anxiety, shame, or cognitive overload of traditional FinOps tools.
Instead of treating cloud cost as a corporate financial audit, CloudSnack treats it as a friendly check-in:
  1. Snackonomics Engine: It translates cold, abstract dollar waste into tangible daily rewards. An idle NAT Gateway isn’t a "$32.40 monthly line item"—it’s "6 oat milk flat whites per month that you could have bought yourself."
  2. Safe Read-Only Cloud Audits: In less than two seconds, it scans your AWS account for the four most common "cloud crumbs": unattached EBS volumes, idle NAT Gateways, unassociated Elastic IPs, and stopped RDS instances that are silently billing storage (or waiting to auto-restart).
  3. Mercy Mode: It guarantees that it will never delete or mutate your cloud resources automatically. Instead, for every single finding, it serves a tested, 1-click copy-paste AWS CLI remediation command pre-loaded with --dry-run or pre-delete snapshot backups.
  4. Cryptic AWS Error Translator: Hit an intimidating error like ClientError: (InvalidParameterCombination) or (ResourceInUseException): The vpc has dependencies? CloudSnack translates it into plain, reassuring English, reassures you that 99% of developers hit this exact wall, and hands you the 30-second fix.

Who It Is For

CloudSnack is built specifically for:
  • Solo Builders, Indie Hackers, and Freelancers who love building on AWS but don't have a dedicated DevOps or FinOps team to watch their back.
  • Hackathon Participants & Students who are experimenting rapidly and frequently forget to tear down temporary infrastructure.
  • Startups & Small Engineering Teams who want to keep cloud burn low without turning their standups into stressful cost-cutting interrogations.

2. How I Built It

To fulfill the challenge ethos—small and delightful beats big and clunky—I avoided bulky heavyweight frameworks that introduce dozens of fragile dependencies. CloudSnack is designed with a modular architecture that combines native AWS SDK primitives with the cutting-edge Amazon Bedrock Converse API.
🏛️ Architecture
CloudSnack Flowchart

AWS Services & Tools Used:

  1. Amazon Bedrock (Converse API): Bedrock serves as the conversational brain of CloudSnack. Using the unified Bedrock Converse API (bedrock-runtime.converse), the agent leverages Anthropic Claude 3.5 Sonnet and Amazon Nova to maintain a warm, witty persona, reason about complex multi-resource relationships (e.g., whether deleting a NAT Gateway will orphan a private subnet's routing table), and answer free-form architectural questions.
  2. AWS SDK for Python (Boto3): Powers the safe, non-destructive scanners. The scanner explicitly calls only describe_* endpoints (describe_volumes, describe_nat_gateways, describe_addresses, describe_db_instances). It contains zero mutation permissions.
  3. Snackonomics Math Engine: A deterministic conversion module that calculates hourly and monthly run rates across AWS resource types and maps them dynamically to local café items, bakery goods, and street food.
  4. Zero-Friction Fallback & Simulator Engine: To ensure that any developer can evaluate CloudSnack within 5 seconds—even before configuring IAM credentials or when operating offline—the codebase includes an intelligent local simulator that mirrors the exact Bedrock agent experience.

The Bedrock Prompting Architecture

The secret to making an agent truly enjoyable lies in tight guardrails around tone and persona. Here is how CloudSnack's system prompt is configured for Amazon Bedrock:
1
2
3
4
5
6
7
8
9
10
SYSTEM_PROMPT = """You are CloudSnack, a warm, witty, and deeply empathetic AI cloud hygiene companion.
Your mission is to help developers, solo builders, and startup teams manage their AWS resources without anxiety, shame, or surprise bills.

Key Personality Guidelines:
1. 'Snackonomics': Whenever discussing dollar costs, convert them into delicious, tangible snack units (oat milk lattes, pizza slices, croissants, street tacos, boba).
2. 'Zero Shame & Empathy': Never scold a user for leaving a resource running. Acknowledge that cloud development is complex and everyone makes mistakes.
3. 'Mercy Mode': Always prioritize safe, non-destructive, reversible actions. Suggest dry-run commands, pre-delete snapshots, and clear rollbacks.
4. 'Plain English': Decrypt acronyms (EBS, EIP, ENI, NAT-GW) into plain human language.
5. Be concise, punchy, cheerful, and actionable.
"""

3. The Delightful Detail (And How I Know It Worked)

If there is one thing that ruins the user experience of AI agents today, it is opaque, judgmental anxiety. When an agent acts like a strict compliance officer or spins invisibly behind a spinner for 20 seconds, people close the tab.
The Delightful Detail in CloudSnack is the combination of "Snackonomics" and "Mercy Mode":

1. Reframing Shame as Reward (Snackonomics)

Most monitoring bots shame you: "You are wasting $55.55/month!"
CloudSnack reframes that exact metric into an empowering reward:
"Whoa, gourmet banquet alert! You have 10.6 Oat Milk Flat Whites ☕ ($55.55/mo) hiding in idle resources. Let's reclaim that coffee budget together."
Suddenly, cleaning up your cloud isn't a tedious chore—it's a scavenger hunt where every clean-up earns you a real-world treat.

2. Mercy Mode: The Anti-Panic Button

Developers hesitate to delete idle cloud resources because they fear the catastrophic mistake: "What if this unattached EBS volume had old database seeds I need?"
Every finding in CloudSnack includes a Mercy Mode Recipe: a two-step snippet that creates a lightweight, cheap snapshot before deletion:
1
2
3
4
5
# 1. Take a safe snapshot just in case (stored in S3 for pennies):
aws ec2 create-snapshot --volume-id vol-03b9841f17e08ba45 --description 'Pre-cleanup safety backup'

# 2. Safely delete the orphaned volume:
aws ec2 delete-volume --volume-id vol-03b9841f17e08ba45

How I Know It Worked

I tested CloudSnack with two fellow developers who had inactive AWS test accounts. Both had previously ignored AWS billing reminder emails for over two months because they were dreading logging into the console.
When they ran python3 run_demo.py:
  • Immediate Laughter: Within 3 seconds of seeing the ASCII banner and reading that their abandoned NAT Gateway had "eaten 6 flat whites while they slept," their posture relaxed completely.
  • Immediate Action: Because the CLI commands were right there with pre-delete snapshot instructions, they ran the remediation recipes on the spot. Both cleared over $40/month in idle waste in less than 5 minutes.
  • The Verdict: One of them remarked: "This is the first time an AWS tool made me smile instead of panic." That was the ultimate confirmation that the detail worked.

4. Proof It Works

Below is the verified, live terminal execution trace of CloudSnack in action:

Step 1: The Welcome Banner & Today's Snack Tray

Running python3 run_demo.py or python3 -m cloudsnack.cli scan:

Step 2: Amazon Bedrock Conversational Guidance

Asking CloudSnack about architectural trade-offs:
1
2
3
4
5
6
7
8
9
10
11
12
13
User > Why is my NAT Gateway costing so much and what is the safest way to replace it?

CloudSnack >
☕ Ah, the infamous NAT Gateway! AWS charges roughly $32.40/month baseline ($0.045/hr)
just to keep it alive—even if zero bytes pass through it. That's about 6 oat milk flat whites!

💡 Mercy Mode Recommendation:
1. If your private EC2 instances only need to talk to S3 or DynamoDB, replace the NAT Gateway
with free VPC Gateway Endpoints.
2. If you don't need outbound internet right now, delete the NAT Gateway:
aws ec2 delete-nat-gateway --nat-gateway-id <ID>

You just bought yourself a weekly coffee budget back!

Step 3: The Cryptic Error Translator

Deciphering real-world AWS error messages (python3 -m cloudsnack.cli translate "..."):
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
┌─────────────────────────────────────────────────────────────┐
│ 🔍 AWS ERROR TRANSLATOR (Plain English + Empathy) │
└─────────────────────────────────────────────────────────────┘

💥 What AWS Said: ClientError: An error occurred (InvalidParameterCombination) when calling CreateDBInstance: AllocatedStorage 10 is less than minimum 20 for storage type gp3

🏷️ Category: RDS Database (RDS Storage Parameter Mismatch)
🗣️ What It Actually Means:
You asked AWS to create an RDS database with storage settings that contradict each other (e.g. asking for gp3 with IOPS lower than 3000 or below the minimum volume size for that engine).

💖 The Empathy Check:
Don't sweat it. AWS parameter combinations have more fine print than a loan agreement. Every senior engineer has wrestled with this exact error.

⚡ Safe Quick Fix (Copy-Paste):
aws rds create-db-instance \
--db-instance-identifier my-db \
--db-instance-class db.t4g.micro \
--engine postgres \
--allocated-storage 20 \
--storage-type gp3

🛡️ Preventative Wisdom:
For dev testing, always stick to 20GB gp3 with default baseline IOPS (3000) so you don't trigger pricing or ratio limits.

Step 4: Automated Test Suite (100% Pass)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
$ python3 -m unittest discover -s tests -v
test_agent_integration (test_cloudsnack.TestCloudSnack.test_agent_integration) ... ok
test_bedrock_converse_simulator (test_cloudsnack.TestCloudSnack.test_bedrock_converse_simulator) ... ok
test_error_translator_general_fallback (test_cloudsnack.TestCloudSnack.test_error_translator_general_fallback) ... ok
test_error_translator_s3_bucket_not_empty (test_cloudsnack.TestCloudSnack.test_error_translator_s3_bucket_not_empty) ... ok
test_error_translator_vpc_dependency (test_cloudsnack.TestCloudSnack.test_error_translator_vpc_dependency) ... ok
test_scanner_simulator (test_cloudsnack.TestCloudSnack.test_scanner_simulator) ... ok
test_snackonomics_positive (test_cloudsnack.TestCloudSnack.test_snackonomics_positive) ... ok
test_snackonomics_zero (test_cloudsnack.TestCloudSnack.test_snackonomics_zero) ... ok

----------------------------------------------------------------------
Ran 8 tests in 0.001s

OK

What I Learned & What's Next

Building an agent that people actually enjoy using requires an inversion of traditional engineering habits:
  1. Tone is as important as accuracy: A hyper-accurate agent that talks like a bureaucrat creates cognitive fatigue. An agent that validates the user's feelings and speaks like a supportive pair-programmer creates genuine affection.
  2. Deterministic safety beats autonomous bravery: Users do not want agents to blindly run delete-volume on their cloud accounts. True delight comes from giving the user superpowers with safety rails, not taking the wheel away from them.
  3. Small and delightful really does beat big and clunky: By focusing on the exact micro-interactions that cause anxiety (surprise bills and cryptic errors), CloudSnack delivers disproportionate joy in under 200 lines of core logic.
Next on the roadmap:
  • A serverless AWS Lambda + Amazon EventBridge cron trigger that drops a weekly "Monday Morning Snack Tray" into your private Slack or Discord channel.
  • Direct integration with AWS Cost Anomaly Detection webhooks.
Try out CloudSnack , reclaim your snack budget, and may your AWS bills stay crisp and crumb-free! 🍪☕
# agents

Series: AI (1 article)

  1. 1
    CloudSnack — The Guilt-Free AI Cloud Hygiene Companion People Actually Enjoy Using This article
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article