Skip to main content

AI Features Documentation - KillIT v3

Overview​

This document provides a comprehensive overview of all AI-powered features in the KillIT v3 application, including AWS Claude (via Bedrock) integrations and use cases.

Table of Contents​

  1. Core AI Services
  2. AI Use Cases
  3. API Endpoints
  4. Configuration
  5. Implementation Examples

Core AI Services​

1. Claude AI Integration (AWS Bedrock)​

Location: /backend/services/ai/claude-service.js

Features:

  • Multiple Claude model support (Claude 3.7 Sonnet, Claude 3.5 Sonnet, Claude Haiku)
  • Rate limiting (30 requests/minute)
  • Automatic retry with exponential backoff
  • Token usage tracking
  • Response validation

Example Usage:

const claudeService = require('./services/ai/claude-service');

// Generate CI insights
const insights = await claudeService.generateResponse(prompt, {
  maxTokens: 2000,
  temperature: 0.7
});

2. Embedding Service (AWS Titan)​

Location: /backend/services/ai/embedding-service.js

Features:

  • Vector embeddings using AWS Titan
  • Batch processing capabilities
  • Redis caching
  • Progress tracking

Use Cases:

  • CI semantic search
  • Relationship similarity matching
  • Knowledge base vectorization

3. RAG (Retrieval-Augmented Generation) Service​

Location: /backend/services/ai/rag-service.js

Features:

  • Hybrid search (vector + keyword)
  • Context-aware responses
  • Entity extraction
  • Relationship queries

AI Use Cases​

Purpose: Natural language search for Configuration Items

How it works:

  1. User enters natural language query (e.g., "Find all Windows servers in production")
  2. Query is converted to embeddings
  3. Vector search finds similar CIs
  4. Results are ranked and returned

API Endpoint: POST /api/cmdb-ai/search

Example Request:

{
  "query": "Show me all database servers with high CPU usage",
  "limit": 10,
  "filters": {
    "type": "Server",
    "status": "active"
  }
}

2. AI-Powered Relationship Discovery​

Purpose: Automatically discover relationships between CIs based on network connections

How it works:

  1. Analyzes network connection data from scan results
  2. Uses Claude to identify relationship patterns
  3. Creates AI-discovered relationships with confidence scores
  4. Learns from user feedback

API Endpoint: POST /api/discovery/process-scan/:scanId

Features:

  • Network topology analysis
  • Service dependency mapping
  • Communication pattern recognition
  • Confidence scoring

3. CI Insights Generation​

Purpose: Generate intelligent insights about CIs

How it works:

  1. Analyzes CI attributes, relationships, and historical data
  2. Uses Claude to generate insights
  3. Identifies risks, dependencies, and recommendations

API Endpoint: POST /api/cmdb-ai/insights/:ciId

Example Insights:

  • "This server hosts 5 critical applications and has no redundancy"
  • "Database server showing unusual network traffic patterns"
  • "Missing security patches for 3 months"

4. Software Classification & Mapping​

Purpose: Automatically classify and map software instances

How it works:

  1. Analyzes software names, versions, and processes
  2. Uses AI to determine software families
  3. Maps to CPE (Common Platform Enumeration)
  4. Creates hierarchical relationships

Features:

  • Automatic vendor identification
  • Version normalization
  • Process-to-application mapping
  • Software family grouping

5. Natural Language Report Generation​

Purpose: Convert natural language queries to complex reports

How it works:

  1. User enters query like "Show me all servers installed last month"
  2. AI converts to MongoDB aggregation pipeline
  3. Executes query and formats results

API Endpoint: POST /api/reports/ai/generate

Example Queries:

  • "List all Windows servers with SQL Server installed"
  • "Show compliance status by department"
  • "Find all software without valid licenses"

6. Compliance Analysis​

Purpose: AI-powered compliance checking

How it works:

  1. Analyzes CI configurations against compliance standards
  2. Identifies gaps and risks
  3. Provides remediation recommendations

Features:

  • NIST, ISO, CIS benchmark checking
  • Custom policy support
  • Risk scoring
  • Automated remediation suggestions

7. Holiday Calendar Generation​

Purpose: Generate holiday calendars using AI

How it works:

  1. Uses OpenAI to generate country-specific holidays
  2. Includes cultural and regional holidays
  3. Fallback to predefined holidays

API Endpoint: POST /api/holidays/generate

API Endpoints​

CMDB AI Endpoints​

  • POST /api/cmdb-ai/search - Semantic search for CIs
  • POST /api/cmdb-ai/find-related - Find related CIs
  • POST /api/cmdb-ai/insights/:ciId - Generate CI insights
  • GET /api/cmdb-ai/embeddings/status - Check embedding status
  • POST /api/cmdb-ai/embeddings/generate - Generate embeddings
  • GET /api/cmdb-ai/test - Test AI connectivity

AI Analytics Endpoints​

  • GET /api/ai-analytics/dashboard - AI usage dashboard
  • GET /api/ai-analytics/queries - Query performance metrics
  • POST /api/ai-analytics/feedback - Submit query feedback
  • GET /api/ai-analytics/optimization-recommendations - Get optimization tips

AI Relationship Endpoints​

  • GET /api/ci/:ciId/ai-relationships - Get AI-discovered relationships
  • POST /api/ai-relationships/analyze - Analyze potential relationships
  • PUT /api/ai-relationships/:id/confidence - Update confidence score
  • POST /api/ai-relationships/feedback - Provide feedback

Configuration​

Environment Variables​

# AWS Configuration
AWS_ACCESS_KEY_ID=your-access-key
AWS_SECRET_ACCESS_KEY=your-secret-key
AWS_REGION=eu-central-1

# Claude Model Selection
CLAUDE_MODEL=eu.anthropic.claude-sonnet-4-5-20250929-v1:0

# OpenAI (for holiday generator)
OPENAI_API_KEY=your-openai-key

# Redis (for caching)
REDIS_URL=redis://localhost:6379

Model Options

// Available Claude models
const CLAUDE_MODELS = {
  CLAUDE_4_5_SONNET: 'eu.anthropic.claude-sonnet-4-5-20250929-v1:0', // Latest, highest quality
  CLAUDE_4_5_HAIKU: 'eu.anthropic.claude-haiku-4-5-20251001-v1:0',   // Fast, cost-effective
  CLAUDE_3_5_SONNET: 'anthropic.claude-3-5-sonnet-20241022-v2:0', // Standard
  CLAUDE_3_HAIKU: 'anthropic.claude-3-haiku-20240307-v1:0'      // Fallback
};

// Embedding model
const EMBEDDING_MODEL = 'amazon.titan-embed-text-v2:0';

Implementation Examples​

// Example implementation
async function getAndStoreInsights(ciId, prompt) {
  const insights = await claudeService.generateResponse(prompt, {
    temperature: 0.7,
    maxTokens: 2000,
    model: 'claude-4.5-sonnet'
  });
  
  // Store insights
  await AIInsights.create({
    ciId,
    insights: insights.content,
    generatedAt: new Date(),
    model: 'claude-4.5-sonnet'
  });
  
  return insights;
}

Best Practices​

  1. Rate Limiting: Always respect rate limits to avoid service interruptions
  2. Caching: Use Redis caching for embeddings to reduce API calls
  3. Error Handling: Implement robust error handling with retries
  4. Feedback Loop: Collect user feedback to improve AI accuracy
  5. Cost Management: Monitor token usage and optimize prompts
  6. Security: Never expose API keys in frontend code
  7. Testing: Use test endpoints to verify AI connectivity

Monitoring & Analytics​

The application includes comprehensive AI monitoring:

  • Token usage tracking
  • Query performance metrics
  • Error rate monitoring
  • User feedback analysis
  • Cost tracking by model

Access the AI Analytics Dashboard at /api/ai-analytics/dashboard for real-time metrics.

ITSM Integrations​

AI Change Manager for Xurrent (4me)​

Location: /backend/controllers/integrations/xurrentWebhookController.js Service: /backend/services/ai/changeRiskAnalysisService.js

Overview: The AI Change Manager provides automated, intelligent risk assessment for IT changes managed in Xurrent (4me) ITSM platform. It leverages NopeSight's CMDB and AI-enhanced relationship data to provide comprehensive change impact analysis.

Architecture:

  1. Webhook Receiver: Accepts webhooks from Xurrent automation rules
  2. AI Analysis Engine: Processes change requests using Claude AI
  3. Risk Calculator: Multi-dimensional risk scoring algorithm
  4. Xurrent Updater: Updates task fields via GraphQL API

Key Features:

1. Multi-Dimensional Risk Assessment​

const riskDimensions = {
  technical: {
    weight: 0.3,
    factors: ['complexity', 'compatibility', 'testing_coverage']
  },
  business: {
    weight: 0.35,
    factors: ['revenue_impact', 'user_count', 'criticality']
  },
  dependency: {
    weight: 0.25,
    factors: ['connected_systems', 'integration_points', 'cascade_potential']
  },
  historical: {
    weight: 0.1,
    factors: ['past_failures', 'change_frequency', 'team_experience']
  }
};

2. Impact Analysis​

  • System Discovery: Identifies all affected systems using CMDB relationships
  • User Impact: Calculates number of affected users
  • Service Mapping: Maps impact to business services
  • Downtime Estimation: Provides risk-adjusted time estimates

3. Intelligent Recommendations​

The AI generates three categories of recommendations:

  • Pre-Change: Validation and preparation steps
  • During-Change: Monitoring and checkpoints
  • Post-Change: Verification and rollback criteria

4. CAB Documentation​

Automatically generates:

  • Executive summary
  • Risk assessment matrix
  • Impact visualization
  • Talking points for CAB meetings
  • Success criteria

API Endpoints:

// Webhook receiver
POST /api/integrations/xurrent/webhook/:configId/change-analysis

// Sample webhook payload
{
  "event": "automation_rule",
  "payload": {
    "ci_id": ["ci-123", "ci-456"],
    "ci_name": ["Database Server", "App Server"],
    "change_note": ["Upgrade to latest version"],
    "task_node_id": "task-789",
    "change_fields": {
      "priority": "high",
      "category": "software_update"
    }
  }
}

Configuration Requirements:

  1. Xurrent API Token Scopes:

    • Automation Rule: Create, Read, Update, Delete
    • Task: Create, Read, Update
    • Task Template: Create, Read, Update
    • UI Extension: Create, Read, Update
    • Webhook: Create, Read, Update, Delete
    • Note: Create, Read

  2. CMDB Prerequisites:

    • Populated CI database
    • Network relationships discovered
    • Software dependencies mapped
    • AI-enhanced relationships (optional but recommended)

Custom Fields in Xurrent: The integration creates 19 custom fields in Xurrent tasks:

Field CategoryFieldsDescription
Risk Scoresai_risk_level, ai_risk_score, ai_technical_risk, ai_business_risk, ai_dependency_risk, ai_historical_riskMulti-dimensional risk assessment
Impact Metricsai_affected_systems, ai_critical_impacts, ai_affected_users, ai_affected_servicesQuantified impact analysis
Time Estimatesai_planned_duration, ai_risk_adjusted_time, ai_service_downtime, ai_recommended_windowDowntime predictions
Recommendationsai_pre_change_rec, ai_during_change_rec, ai_post_change_recActionable guidance
Analysisai_summary, ai_insights, ai_confidenceAI analysis metadata

Implementation Example:

// Process change webhook
async function processChangeAnalysis(webhookData, configId) {
  // 1. Extract CI information
  const ciIds = webhookData.ci_id;
  const changeNote = webhookData.change_note;

  // 2. Fetch CI details from CMDB
  const cis = await CI.find({
    sourceID: { $in: ciIds }
  }).populate('relationships');

  // 3. Analyze with AI
  const analysis = await changeRiskAnalysisService.analyzeChange({
    ciId: cis[0]._id,
    changeDescription: changeNote,
    changeType: webhookData.change_fields?.category
  });

  // 4. Update Xurrent task
  await xurrentService.updateTaskWithAIAnalysis(
    webhookData.task_node_id,
    analysis
  );
}

Performance Metrics:

  • Average analysis time: 5-10 seconds
  • Webhook response time: <200ms (async processing)
  • Field update time: 2-3 seconds
  • Success rate: >98%

Best Practices:

  1. Ensure CMDB is regularly updated with discovery scans
  2. Map critical systems for accurate impact assessment
  3. Configure product mappings for better Xurrent integration
  4. Monitor webhook delivery status
  5. Review AI confidence scores for critical changes

Future Enhancements​

  1. Predictive Maintenance: Use historical data to predict failures
  2. Anomaly Detection: Real-time anomaly detection in CI behavior
  3. Automated Remediation: AI-driven problem resolution
  4. Natural Language CI Updates: Update CIs using natural language
  5. Multi-modal Analysis: Incorporate logs and metrics in analysis
  6. Enhanced Change Manager:
    • Learning from change outcomes
    • Automated change scheduling
    • Rollback automation triggers
    • Cross-platform ITSM support