Skip to content

CLAUDE-SETUP.md

Latest commit

 

History

History
186 lines (147 loc) · 4.63 KB

CLAUDE-SETUP.md

File metadata and controls

186 lines (147 loc) · 4.63 KB

Claude AI Setup for DevOps Examples

Quick Start

All DevOps examples now use Claude AI by default for intelligent analysis. Follow these steps:

1. Get Your Claude API Key

  1. Go to: https://console.anthropic.com/settings/keys
  2. Create a new API key
  3. Copy the key (starts with sk-ant-)

2. Set Up Environment

# Option A: Create .env file (recommended)
cp .env.example .env
# Edit .env and add your CLAUDE_API_KEY

# Option B: Export directly
export CLAUDE_API_KEY="sk-ant-your-key-here"
export CLAUDE_DEBUG_LOGGING=true  # To see all prompts/responses

3. Run Examples

# Using the run script (recommended - handles everything)
cd cost-optimizer
./run.sh

# Or run directly with environment
source ../setup-env.sh
./cost-optimizer

Features When Claude is Enabled

Cost Optimizer

  • Intelligent cost analysis based on actual usage patterns
  • Risk assessment for each recommendation (low/medium/high)
  • Specific optimization actions with expected savings
  • Context-aware suggestions based on your infrastructure

Drift Detector

  • Root cause analysis of configuration drift
  • Automated fix recommendations with safety ratings
  • Pattern recognition for recurring drift issues

How to Disable Claude

Method 1: Environment Variable (Temporary)

# Disable for one run
ENABLE_CLAUDE=false ./run.sh

# Or export for session
export ENABLE_CLAUDE=false
./cost-optimizer

Method 2: Edit .env File (Persistent)

# In .env file, change:
ENABLE_CLAUDE=false

Method 3: Comment Out in Code

// In main.go, comment out Claude initialization:
// app.Claude = sdk.NewClaudeClient(os.Getenv("CLAUDE_API_KEY"))

Debug Logging

Claude debug logging is enabled by default to show all prompts and responses.

View Full Prompts and Responses

export CLAUDE_DEBUG_LOGGING=true  # Already default
./cost-optimizer

Output will show:

[Claude] req-1 ◀ FULL_PROMPT:
Analyze the following Kubernetes resource usage...

[Claude] req-1 ▶ FULL_RESPONSE:
{
  "total_monthly_cost": 922.00,
  "recommendations": [...]
}

Disable Debug Logging

export CLAUDE_DEBUG_LOGGING=false

What Claude Logs

When debug logging is enabled, you'll see:

  1. Request Counter: req-1, req-2, etc.
  2. Full Prompts: Complete text sent to Claude
  3. Full Responses: Complete JSON/text returned
  4. Timing: Duration of each API call
  5. Errors: Any API failures with details

Example log output:

[Claude] 🔍 Debug logging enabled - all prompts and responses will be logged
[Claude] req-1 ◀ REQUEST: Analyze the following Kubernetes resource usage data...
[Claude] req-1 ◀ FULL_PROMPT:
Analyze the following Kubernetes resource usage data and provide cost optimization recommendations.
Focus on:
1. Resources with low utilization (<50%)...

[Claude] req-1 → Sending API request
[Claude] req-1 ▶ RESPONSE (1.2s): {"total_monthly_cost": 922.00, "potential_savings": 208.95...
[Claude] req-1 ▶ FULL_RESPONSE:
{
  "total_monthly_cost": 922.00,
  "potential_savings": 208.95,
  "savings_percentage": 22.8,
  "recommendations": [
    {
      "resource": "deployment/frontend-web",
      "priority": "high",
      "monthly_savings": 73.65,
      "risk": "low",
      "explanation": "Frontend is over-provisioned..."
    }
  ]
}

ConfigHub Integration

ConfigHub is also enabled by default. Authentication happens automatically:

# Token is obtained automatically from cub CLI
# Or set manually:
export CUB_TOKEN=$(cub auth get-token)

Standard Pattern for All Examples

Every example follows this pattern:

  1. run.sh script - Handles all setup automatically
  2. Claude enabled by default - With easy disable option
  3. Debug logging by default - See all AI interactions
  4. ConfigHub integration - Automatic authentication
  5. Fallback to basic - Works without Claude if needed

Troubleshooting

"Claude API key not set"

# Check your key is exported
echo $CLAUDE_API_KEY

# Or add to .env file
echo "CLAUDE_API_KEY=sk-ant-your-key" >> .env

"API error 401: Unauthorized"

Your API key is invalid. Get a new one from https://console.anthropic.com/settings/keys

"Claude analysis failed"

Check debug logs for details:

export CLAUDE_DEBUG_LOGGING=true
./run.sh

Cost Considerations

  • Claude API calls cost ~$0.25 per 1M input tokens
  • Each analysis typically uses ~2-5K tokens
  • Cost optimizer runs analysis every 30 seconds
  • Approximate cost: $0.01-0.02 per hour of operation

To reduce costs:

  • Increase analysis interval in code
  • Use ENABLE_CLAUDE=false for development
  • Enable only for production monitoring