Infrastructure Cost Estimation

Architecture: Skill Orchestrates, Agent Calculates

┌─────────────────────────────────────────────────────────────┐
│                    SKILL (Orchestrator)                     │
│                                                             │
│  Step 1: Select Products                                    │
│    - Access Manager: ALWAYS (shared platform)               │
│    - Midaz Core: [YES / NO]                                 │
│    - Reporter: [YES / NO]                                   │
│                                                             │
│  Step 2: Basic Info                                         │
│    - Repo path, Total customers                             │
│                                                             │
│  Step 2a: Infrastructure Sizing (NEW in v6.0)               │
│    - Option 1: Tier (Starter/Growth/Business/Enterprise)    │
│    - Option 2: Custom TPS (backwards compatible)            │
│                                                             │
│  Step 3: Select Environment(s) to Calculate                 │
│    - [x] Homolog (us-east-2, Single-AZ, 1 replica)         │
│    - [x] Production (sa-east-1, Multi-AZ, 3 replicas)      │
│                                                             │
│  Step 4: Read Helm Charts (for selected products only)      │
│    - ALWAYS: charts/plugin-access-manager/values.yaml       │
│    - If Midaz: charts/midaz/values.yaml                     │
│    - If Reporter: charts/reporter/values.yaml               │
│                                                             │
│  Step 5: Ask PER COMPONENT: Shared or Dedicated?            │
│    - VPC, EKS, PostgreSQL, Valkey, etc.                     │
│                                                             │
│  Steps 6-7: Database Config + Billing Model                 │
│                                                             │
│  ↓ All data collected (products + tier/TPS + Helm configs)  │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                    AGENT (Calculator)                       │
│                                                             │
│  Receives: Products + Tier/TPS + Helm configs → Calculates: │
│  - Infrastructure costs PER ENVIRONMENT (Homolog + Prod)    │
│  - EKS node sizing from tier config OR actual CPU/memory    │
│  - Cost attribution (shared ÷ customers, dedicated = full)  │
│  - Access Manager costs ALWAYS shared across ALL customers  │
│  - Profitability analysis (using combined env costs)        │
│                                                             │
│  Returns: Side-by-side Homolog vs Production breakdown      │
└─────────────────────────────────────────────────────────────┘

Step 1: Select Products for This Customer

Ask which products the customer needs:

Use AskUserQuestion for product selection:

AskUserQuestion:
  question: "Which products does this customer need? (Access Manager is always included)"
  header: "Products"
  multiSelect: true
  options:
    - label: "Midaz Core + Reporter (Recommended)"
      description: "Full platform: ledger + regulatory reporting"
    - label: "Midaz Core only"
      description: "Base ledger platform without reporting"

Product → Helm Chart Mapping:

Step 2: Gather Basic Information

Why Helm Charts Repo? LerianStudio/helm contains actual CPU/memory configurations per service. Without it, the agent uses Midaz default values.

Note: TPS/Tier selection moved to Step 2a below.

Step 2a: Select Infrastructure Sizing Method (NEW in v6.0)

Choose between pre-configured tier or custom TPS:

2a.1: Ask Sizing Method

AskUserQuestion:
  question: "How would you like to size the infrastructure?"
  header: "Sizing Method"
  multiSelect: false
  options:
    - label: "Pre-configured Tier (Recommended)"
      description: "Choose optimized tier by TPS range - faster, cost-optimized"
    - label: "Custom TPS"
      description: "Specify exact TPS for custom infrastructure sizing"

2a.2: If “Pre-configured Tier” Selected

AskUserQuestion:
  question: "Which infrastructure tier matches your expected traffic?"
  header: "Tier"
  multiSelect: false
  options:
    - label: "Starter (50-100 TPS) - ~R$ 11.7k/mo"
      description: "Small deployments, POC, single tenant. 2× c6i.large nodes."
    - label: "Growth (100-500 TPS) - ~R$ 19.3k/mo"
      description: "Growing business, moderate traffic. 3× c6i.xlarge nodes."
    - label: "Business (500-1,500 TPS) - ~R$ 27.5k/mo (Recommended)"
      description: "Production app, full HA. 4× c6i.xlarge nodes, Multi-AZ."
    - label: "Enterprise (1,500-5,000 TPS) - ~R$ 56.7k/mo"
      description: "Mission critical, high performance. 8× c6i.2xlarge nodes."

Tier Reference: See infrastructure-tiers-by-tps.md for detailed tier specifications.

Tier Configurations (stored for agent dispatch):

2a.3: If “Custom TPS” Selected

AskUserQuestion:
  question: "What is the expected TPS (transactions per second)?"
  header: "TPS"
  multiSelect: false
  options:
    - label: "50 TPS"
      description: "Low traffic, testing"
    - label: "150 TPS"
      description: "Small to medium traffic"
    - label: "500 TPS"
      description: "Medium to high traffic"
    - label: "Custom value"
      description: "I'll specify exact TPS"

Custom TPS Calculation:

• Agent will calculate required replicas: (TPS ÷ 670) × 1.25 for auth service
• Agent will size EKS nodes based on total CPU/memory requirements
• More flexible but requires more calculation

Step 3: Select Environment(s) to Calculate

Ask which environments need cost estimation:

Use AskUserQuestion for environment selection:

AskUserQuestion:
  question: "Which environments should be estimated?"
  header: "Environments"
  multiSelect: true
  options:
    - label: "Both (Recommended)"
      description: "Calculate Homolog + Production costs for complete picture"
    - label: "Production only"
      description: "Calculate only production environment (São Paulo)"
    - label: "Homolog only"
      description: "Calculate only homolog/staging environment (Ohio)"

Environment Differences:

Step 4: Read Selected Product Helm Charts

4a. Always Read Access Manager (ALWAYS SHARED)

Access Manager is platform-level infrastructure – ALWAYS included, ALWAYS shared across all customers.

ALWAYS READ: charts/plugin-access-manager/values.yaml

Services:
- identity (100m CPU, 128Mi memory)
- auth (500m CPU, 256Mi memory)

Infrastructure (shared):
- auth-database (PostgreSQL for Casdoor)
- valkey (session cache)

Cost Attribution: Access Manager costs are ALWAYS divided by total platform customers.

4b. Read Selected Products (Based on Step 1)

Read ONLY the Helm charts for products selected in Step 1:

Example – Customer selected “Midaz Core + Reporter”:

Read: charts/plugin-access-manager/values.yaml  → ALWAYS (shared platform)
Read: charts/midaz/values.yaml                  → Selected
Read: charts/reporter/values.yaml               → Selected

Example – Customer selected “Midaz Core only”:

Read: charts/plugin-access-manager/values.yaml  → ALWAYS (shared platform)
Read: charts/midaz/values.yaml                  → Selected
Skip: charts/reporter/values.yaml               → Not selected

4c. Extract Resource Configurations

Source: git@github.com:LerianStudio/helm.git

For each service, extract:

resources:
  requests:
    cpu: ???m      # CPU request in millicores
    memory: ???Mi  # Memory request
autoscaling:
  minReplicas: ?   # Minimum replicas
  maxReplicas: ?   # Maximum replicas

Example from LerianStudio/helm values.yaml:

transaction:
  replicaCount: 3
  resources:
    requests:
      cpu: 2000m
      memory: 512Mi
    limits:
      cpu: 2000m
      memory: 512Mi
  autoscaling:
    enabled: true
    minReplicas: 3
    maxReplicas: 9

If no Helm repo available, use Midaz default values from this document.

4d. Read Actual Resources (DYNAMIC – from LerianStudio/helm)

MUST read actual values at runtime – DO NOT use hardcoded values.

Source Repository: git@github.com:LerianStudio/helm.git

Files to Read:

What to Extract per Service:

# For each service, extract:
resources:
  requests:
    cpu: ???m      # CPU request in millicores
    memory: ???Mi  # Memory request
  limits:
    cpu: ???m      # CPU limit
    memory: ???Mi  # Memory limit

autoscaling:
  minReplicas: ?   # Minimum replicas
  maxReplicas: ?   # Maximum replicas

# For databases (postgresql, mongodb, valkey, rabbitmq):
# Look for resourcesPreset or explicit resources block

How to Read:

1. If local clone exists: Read tool on values.yaml files
2. If no local clone: WebFetch from GitHub raw URLs:
   • https://raw.githubusercontent.com/LerianStudio/helm/main/charts/midaz/values.yaml
   • https://raw.githubusercontent.com/LerianStudio/helm/main/charts/reporter/values.yaml
   • https://raw.githubusercontent.com/LerianStudio/helm/main/charts/plugin-access-manager/values.yaml

Fallback: If plugin/service not found, use Midaz core setup as baseline.

Step 5: Ask Per-Component Sharing Model (CRITICAL)

This step requires asking about BOTH database types in a SINGLE question:

• PostgreSQL (RDS) – Relational database
• DocumentDB (MongoDB) – Document database

FORBIDDEN: Asking only about PostgreSQL and forgetting DocumentDB. REQUIRED: Use the multiSelect AskUserQuestion with ALL components listed.

Reference: See infrastructure-cost-estimation-guide.md for:

• Sharing model definitions (SHARED vs DEDICATED vs ALWAYS SHARED)
• How sharing works in practice per component
• NAT Gateway architecture (ALWAYS SHARED)

Components That MUST Be Asked:

1. VPC, 2. EKS Nodes, 3. PostgreSQL, 4. DocumentDB, 5. Valkey, 6. RabbitMQ

Note: NAT Gateway and ALB are ALWAYS SHARED (platform-level resources).

MANDATORY: Use AskUserQuestion tool with multiSelect:

AskUserQuestion:
  question: "Which components need DEDICATED instances? (All others use schema-based sharing)"
  header: "Dedicated"
  multiSelect: true
  options:
    - label: "VPC"
      description: "Fully isolated network (separate VPC per customer)"
    - label: "PostgreSQL"
      description: "Fully isolated RDS instance (not schema-based)"
    - label: "DocumentDB"
      description: "Fully isolated DocumentDB cluster"
    - label: "Valkey"
      description: "Fully isolated ElastiCache cluster"
    - label: "RabbitMQ"
      description: "Fully isolated Amazon MQ broker"
    - label: "EKS Nodes"
      description: "Dedicated compute nodes (separate node group)"

Step 5 Verification Checklist:

[ ] Used multiSelect question (not individual questions per component)
[ ] PostgreSQL included in options
[ ] DocumentDB included in options
[ ] All 6 configurable components listed
[ ] User response captured for ALL components

If any checkbox is NO → STOP and fix before proceeding to Step 6.

Step 6: Database Configuration (Production Only)

For PRODUCTION environment, ask about database HA configuration:

Multi-AZ (High Availability)

Default: Production = Multi-AZ YES, Homolog = Multi-AZ NO

Read Replicas (Scale Reads)

Use AskUserQuestion for database HA:

AskUserQuestion:
  question: "For PRODUCTION databases, what HA configuration?"
  header: "Database HA"
  multiSelect: false
  options:
    - label: "Multi-AZ + Read Replicas (Recommended)"
      description: "Full HA: automatic failover + read scaling"
    - label: "Multi-AZ only"
      description: "Automatic failover, no read replicas"
    - label: "Single-AZ (Not recommended)"
      description: "No HA - only for cost-sensitive scenarios"

Note: Homolog/Staging always uses Single-AZ, no read replicas (testing only).

Step 6b: Backup Configuration (Per Environment)

Backup policies differ significantly between Homolog and Production:

Backup Types

Environment-Specific Defaults

Backup Cost Components

Backup Sizing by TPS

Backup Configuration Questions

Use AskUserQuestion for backup policy:

AskUserQuestion:
  question: "What backup retention policy for PRODUCTION?"
  header: "Backup Policy"
  multiSelect: false
  options:
    - label: "Standard (7-day retention, daily snapshots) (Recommended)"
      description: "7-day PITR + 7 daily snapshots (moderate cost)"
    - label: "Extended (35-day retention, daily + weekly snapshots)"
      description: "35-day PITR + daily/weekly snapshots (higher cost)"
    - label: "Minimal (1-day retention, weekly snapshots)"
      description: "1-day PITR + weekly snapshots (lower cost, higher risk)"

Homolog Backup Policy:

• Always minimal (1-7 day retention)
• No additional snapshots needed (testing environment)
• Cost: Typically free (within RDS automated backup limit)

Step 7: Gather Billing Model

Step 8: Dispatch Agent with Complete Data

BEFORE dispatching, skill MUST read actual resource configs:

8a. Read Current Values from LerianStudio/helm (Based on Selected Products)

# ALWAYS read (platform-level, shared)
Read: charts/plugin-access-manager/values.yaml → Extract: identity, auth resources

# Read based on product selection (Step 1)
If Midaz Core selected:
  Read: charts/midaz/values.yaml → Extract: onboarding, transaction, ledger, crm resources

If Reporter selected:
  Read: charts/reporter/values.yaml → Extract: manager, worker, frontend resources

# Also extract database configurations for selected products

8b. Dispatch Agent with Collected Data

Only dispatch AFTER reading actual values for selected products.

Two dispatch patterns:

1. Tier-based (user selected pre-configured tier)
2. Custom TPS (user specified exact TPS)

Pattern 1: Tier-Based Dispatch

If user selected a tier in Step 2a, provide tier configuration:

Task tool:
  subagent_type: "ring:infrastructure-cost-estimator"
  model: "opus"
  prompt: |
    Calculate infrastructure costs using PRE-CONFIGURED TIER.

    Tier: BUSINESS (500-1,500 TPS)
    Target TPS: 1,500
    Max Capacity: 2,010 TPS

    Tier Configuration:
    Production:
      Replicas: auth=3, transaction=3, onboarding=2, identity=1, auth-backend=1
      Nodes: 4× c6i.xlarge (4 vCPU, 8 GiB each)
      Databases:
        - auth-postgresql: db.m7g.large (Multi-AZ + 1 replica)
        - midaz-postgresql: db.m7g.large (Multi-AZ + 1 replica)
        - documentdb: db.r8g.large (Multi-AZ + 1 replica)
      Cache:
        - auth-valkey: cache.m7g.large (Multi-AZ)
        - midaz-valkey: cache.m7g.large (Multi-AZ)
      Queue:
        - rabbitmq: mq.m7g.large (active/standby)

    Homolog:
      Nodes: 2× c6i.xlarge
      Databases: All db.m7g.large (Single-AZ)

    Products Selected:
    - Access Manager: ALWAYS INCLUDED (shared platform)
    - Midaz Core: YES
    - Reporter: NO

    Infrastructure:
    - App Repo: /workspace/midaz
    - Total Customers on Platform: 5

    Environments to Calculate: [Homolog, Production]

    Actual Resource Configurations (from Helm charts):
    [INSERT VALUES FROM charts/plugin-access-manager/values.yaml]
    [INSERT VALUES FROM charts/midaz/values.yaml]

    Component Sharing Model:
    [... same as before ...]

    Database Configuration:
    [... same as before ...]

    Billing Model:
    [... same as before ...]

    Calculate and return:
    [... same 11 sections as before ...]

Pattern 2: Custom TPS Dispatch

If user specified custom TPS, agent calculates infrastructure:

Task tool:
  subagent_type: "ring:infrastructure-cost-estimator"
  model: "opus"
  prompt: |
    Calculate infrastructure costs for CUSTOM TPS.

    TPS: 750 (custom)

    Calculate required infrastructure:
    - Required auth replicas = (750 ÷ 670) × 1.25 = ~2 replicas
    - Required transaction replicas = (750 ÷ 815) × 1.25 = ~2 replicas
    - Calculate EKS node sizing from total CPU/memory requirements
    - Use appropriate database instance sizes for traffic level

    Products Selected:
    - Access Manager: ALWAYS INCLUDED (shared platform)
    - Midaz Core: YES
    - Reporter: NO

    Infrastructure:
    - App Repo: /workspace/midaz
    - Helm Charts Source: LerianStudio/helm
    - Total Customers on Platform: 5

    Environments to Calculate: [Homolog, Production]

    Actual Resource Configurations (from Helm charts):
    [INSERT VALUES FROM charts/plugin-access-manager/values.yaml]
    [INSERT VALUES FROM charts/midaz/values.yaml]

    Component Sharing Model:
    [... same as before ...]

    Database Configuration:
    [... same as before ...]

    Billing Model:
    [... same as before ...]

    Calculate and return:
    [... same 11 sections as before ...]

Note: Custom TPS dispatch is backwards compatible with v5.0.

Quick Reference

Pricing Tables: See infrastructure-cost-estimation-guide.md for complete AWS pricing (São Paulo and Ohio regions).

Expected Output Sections (from Agent)

Full Output Format: See infrastructure-cost-estimation-guide.md for detailed output section descriptions.

The agent returns 11 required sections:

1. Discovered Services
2. Compute Resources (from LerianStudio/helm)
3. Homolog Environment Costs
4. Production Environment Costs
5. Environment Comparison
6. Infrastructure Components (Consolidated)
7. Cost by Category
8. Shared vs Dedicated Summary
9. TPS Capacity Analysis
10. Profitability Analysis (Combined Environments)
11. Summary

Example Workflow

User Request:

“Estimate costs for Midaz with 100 TPS, 5 customers sharing, PostgreSQL dedicated”

Step 1: Select Products

Products Selected:
- Access Manager: ALWAYS INCLUDED (shared)
- Midaz Core: YES
- Reporter: YES (full platform)

Step 2: Basic Info

Repo: /workspace/midaz
Total Customers: 5

Step 2a: Sizing Method

User selected: "Pre-configured Tier"
Tier: STARTER (50-100 TPS)

Steps 3-7: Skill Gathers Data

Environments: Both (Homolog + Production)

Component Sharing:
- EKS Cluster: SHARED (5)
- PostgreSQL: DEDICATED (1)  ← user specified
- Valkey: SHARED (5)
- DocumentDB: SHARED (5)
- RabbitMQ: SHARED (5)

Billing:
- Unit: transaction
- Price: R$ 0.10
- Volume: 1,000,000/month

Step 8a: Skill Reads LerianStudio/helm (for selected products)

# ALWAYS read (platform-level)
Read: charts/plugin-access-manager/values.yaml
  → identity: 100m CPU, 128Mi memory, 1-3 replicas
  → auth: 500m CPU, 256Mi memory, 3-9 replicas

# Midaz Core (selected)
Read: charts/midaz/values.yaml
  → onboarding: 1500m CPU, 512Mi memory, 2-5 replicas
  → transaction: 2000m CPU, 512Mi memory, 3-9 replicas
  → ledger: 1500m CPU, 256Mi memory, 2-9 replicas

# Reporter (selected)
Read: charts/reporter/values.yaml
  → manager: 100m CPU, 256Mi memory
  → worker: 100m CPU, 128Mi memory
  → frontend: 100m CPU, 128Mi memory

Step 8b: Skill Dispatches Agent

Agent receives: Products selected + actual Helm values + all collected data

Agent Returns:

## Summary

| Metric | Value |
|--------|-------|
| Shared Infrastructure | R$ 1,018/customer |
| Dedicated Infrastructure | R$ 1,490/customer |
| **Total Cost/Customer** | **R$ 2,508/month** |
| Monthly Revenue | R$ 100,000 |
| Gross Profit | R$ 97,492 |
| Gross Margin | 97.5% |

Anti-Rationalization

Checklist Before Dispatch

STEP 1 - PRODUCTS:
[ ] Products selected (Midaz Core, Reporter)?
[ ] Access Manager included (ALWAYS)?

STEP 2 - BASIC INFO:
[ ] Repo path collected?
[ ] Total customers collected?

STEP 2a - SIZING (NEW in v6.0):
[ ] Sizing method selected (Tier OR Custom TPS)?
  If Tier:
    [ ] Tier selected (Starter/Growth/Business/Enterprise)?
  If Custom TPS:
    [ ] TPS value collected?

STEP 3 - ENVIRONMENTS:
[ ] Environments selected (Homolog, Production, Both)?

STEP 4 - HELM CHARTS:
[ ] LerianStudio/helm values read for selected products?

STEP 5 - COMPONENT SHARING (CRITICAL):
DATABASE COMPONENTS - verify BOTH:
[ ] PostgreSQL sharing model collected? (SHARED or DEDICATED)
[ ] DocumentDB sharing model collected? (SHARED or DEDICATED)

OTHER COMPONENTS:
[ ] VPC sharing model collected?
[ ] EKS Nodes sharing model collected?
[ ] Valkey sharing model collected?
[ ] RabbitMQ sharing model collected?

STEP 6-7 - DATABASE & BILLING:
[ ] Database HA configuration collected?
[ ] Backup policy collected?
[ ] Billing unit collected?
[ ] Price per unit collected?
[ ] Expected volume collected?

If any NO → Ask user first, then dispatch.
⛔ If PostgreSQL OR DocumentDB is missing → STOP and ask about BOTH databases.