Slate - Pre-Deployment AWS Cost Estimation for CDK

Hats I Wore

Project Overview

AI-assisted IaC development generates working infrastructure fast — but a single configuration decision (NAT Gateway, multi-AZ RDS, DynamoDB billing mode change) can shift monthly costs by hundreds of dollars silently. Slate makes that visible before you deploy.

How It Works

cdk synth → CDK Adapter → Classifier → AWS Price List API → Free Tier Deduction → Calculator → Output

1. CDK Adapter — reads cdk.out/manifest.json + CloudFormation templates, normalizes resources
2. Classifier — categorizes each resource as Fixed (exact price) or Variable (scales with users)
3. Pricing Client — calls AWS Price List API for region-accurate unit prices (never hardcoded)
4. Free Tier — deducts AWS Free Tier allowances for accurate low-tier estimates
5. Calculator — multiplies variable costs across user growth tiers
6. Output — renders tier table, generates cost-profile.yaml, writes report JSON

Key Features

• Zero config — works immediately after cdk synth with no setup
• Real prices — all costs from AWS Price List API, never hardcoded
• Tiered output — costs across 0–100, 100–1k, 1k–10k, 10k–100k, 100k–1M users
• Free tier modeling — accurate $0 estimates at low usage
• Cost drift detection — diffs between runs show what changed
• CI/CD budget gate — exits code 1 if cost exceeds threshold
• MCP server — expose cost tools to AI editors (Kiro, VS Code, Continue)
• Interactive wizard — guided CLI to fill in usage assumptions

The Precision Dial

cost-profile.yaml is generated with fields for detected variable resources. Fill in what you know — leave blank what you don't.

| Fields filled | Output | |---|---| | None | All variable costs shown as min–max range | | Some | Filled fields exact, empty fields ranged | | All | Every line item exact |

Technologies

TypeScript, ESM, Commander, @aws-sdk/client-pricing, @inquirer/prompts, js-yaml, chalk, tsup, vitest