🤖 AI Capabilities in ANSAI¶

ANSAI integrates AI throughout the automation stack - not just in your IDE, but in the actual infrastructure automation itself.

How AI Works in ANSAI¶

┌─────────────────────────────────────────────────────────────┐
│                    ANSAI AI Integration                      │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  Service Fails → Gather Logs → AI Analysis → Healing Action │
│                      ↓                                       │
│              Root Cause Report                               │
│              + Recommendations                               │
│              + Prevention Tips                               │
│                                                              │
└─────────────────────────────────────────────────────────────┘

What the AI Does¶

When a service fails, ANSAI's self-healing system:

Gathers diagnostic data - systemctl status, recent logs, system resources
Sends to AI for analysis - Groq, OpenAI, Ollama, or any LLM
Gets actionable insights:
ROOT CAUSE: Why did this fail?
WHY IT FAILED: Specific technical details
RECOMMENDED FIX: Commands to run
PREVENTION: How to avoid recurrence
Includes analysis in alerts - Email, webhook, or logs

Example AI Analysis Output¶

══════════════════════════════════════════════════
🤖 AI-POWERED ROOT CAUSE ANALYSIS
══════════════════════════════════════════════════

1. ROOT CAUSE: The service crashed due to memory exhaustion 
   after a memory leak in the connection pool handler.

2. WHY IT FAILED:
   • Memory usage grew from 256MB to 2GB over 6 hours
   • OOM killer terminated the process (signal 9)
   • No memory limits configured in systemd unit

3. RECOMMENDED FIX:
   sudo systemctl edit myapp.service
   # Add: MemoryMax=1G
   sudo systemctl daemon-reload
   sudo systemctl restart myapp

4. PREVENTION: Set memory limits in the service unit file
   and enable automatic restart on OOM.

Setting Up AI¶

Option 1: Groq (Recommended - Fast & Free Tier)¶

# 1. Get free API key at https://console.groq.com
# 2. Export before running playbooks
export ANSAI_GROQ_API_KEY='gsk_...'

# 3. Deploy with AI enabled
cd ~/.ansai/orchestrators/ansible
ansible-playbook playbooks/deploy-self-healing.yml \
  -i inventory/hosts.yml \
  -e 'monitored_services=[{"name": "nginx", "port": 80}]'

Why Groq? - Free tier with generous limits - Fastest inference (llama-3.1-8b-instant) - No credit card required

Option 2: Ollama (100% Local - No API Keys)¶

# 1. Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# 2. Pull a model
ollama pull llama3

# 3. Configure ANSAI to use Ollama
ansible-playbook playbooks/deploy-self-healing.yml \
  -e "ai_backend=ollama" \
  -e "ollama_url=http://localhost:11434" \
  -e "ollama_model=llama3"

Why Ollama? - Completely free and private - No internet required - Data never leaves your machine

Option 3: LiteLLM Proxy (Multi-Model)¶

# 1. Start the proxy
ansai-litellm-proxy

# 2. Configure ANSAI to use it
ansible-playbook playbooks/deploy-self-healing.yml \
  -e "ai_backend=litellm" \
  -e "litellm_url=http://localhost:4000"

Why LiteLLM? - Use any model (OpenAI, Claude, Groq, Ollama) - Automatic fallback between providers - Cost tracking and optimization

AI Tools Included¶

ANSAI includes several AI-powered CLI tools:

ansai-litellm-proxy¶

Multi-model LLM proxy server.

# Start proxy on port 4000
ansai-litellm-proxy

# Use with any OpenAI-compatible client
curl http://localhost:4000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4", "messages": [{"role": "user", "content": "Hello"}]}'

ansai-log-analyzer¶

AI-powered log analysis.

# Analyze recent logs
ansai-log-analyzer /var/log/nginx/error.log

# With time filter
ansai-log-analyzer /var/log/syslog --since "1 hour ago"

ansai-incident-report¶

Generate AI incident reports from logs.

# Generate incident report
ansai-incident-report --service nginx --timeframe "2 hours"

Testing AI Locally¶

Quick Test (No Service Failures Required)¶

# Test Groq API directly
curl -X POST "https://api.groq.com/openai/v1/chat/completions" \
  -H "Authorization: Bearer $ANSAI_GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.1-8b-instant",
    "messages": [{"role": "user", "content": "Why might nginx fail to start?"}]
  }'

Simulate a Failure¶

# 1. Stop a non-critical service
sudo systemctl stop cockpit.socket

# 2. Watch the self-healing logs
journalctl -t ansai-self-heal -f

# 3. Look for AI analysis in the output
# You should see "🤖 AI-POWERED ROOT CAUSE ANALYSIS"

View Historical AI Analyses¶

# All healing events with AI analysis
journalctl -t ansai-self-heal | grep -A 20 "AI-POWERED"

# Recent healing reports
ls -la /tmp/self-heal-*.txt
cat /tmp/self-heal-nginx-*.txt

Configuration Reference¶

defaults/main.yml Options¶

# AI-Powered Root Cause Analysis
ai_analysis_enabled: true          # Enable/disable AI analysis
ai_backend: groq                   # groq, ollama, litellm, openai

# Groq Configuration (default)
groq_api_key: "{{ lookup('env', 'ANSAI_GROQ_API_KEY') }}"
groq_model: llama-3.1-8b-instant
groq_api_url: https://api.groq.com/openai/v1/chat/completions

# Ollama Configuration (local)
ollama_url: http://localhost:11434
ollama_model: llama3

# LiteLLM Configuration (proxy)
litellm_url: http://localhost:4000
litellm_model: gpt-4

Environment Variables¶

Variable	Description	Example
`ANSAI_GROQ_API_KEY`	Groq API key	`gsk_abc123...`
`OPENAI_API_KEY`	OpenAI API key	`sk-abc123...`
`ANTHROPIC_API_KEY`	Claude API key	`sk-ant-...`

Troubleshooting AI¶

"AI analysis unavailable"¶

# Check if API key is set
echo $ANSAI_GROQ_API_KEY

# Test API connectivity
curl -s https://api.groq.com/openai/v1/models \
  -H "Authorization: Bearer $ANSAI_GROQ_API_KEY"

"AI Error: rate limit exceeded"¶

Groq free tier has rate limits
Solution: Use Ollama locally or upgrade Groq plan

AI analysis not appearing in reports¶

# Check if AI is enabled in the deployed scripts
grep "AI_ENABLED" /usr/local/bin/self-heal/*.sh

# Should show: AI_ENABLED="true"

ansai-ask: Interactive AI Assistant¶

Chat with AI about your infrastructure directly from the command line.

Basic Usage¶

# Ask a question
ansai-ask "why might nginx fail to start?"

# Analyze a log file
ansai-ask --analyze /var/log/nginx/error.log

# Ask about a specific service
ansai-ask --service nginx "why is it using so much memory?"

# Pipe logs for analysis  
journalctl -u myapp --since "1 hour ago" | ansai-ask --stdin "what's causing these errors?"

# Interactive chat mode
ansai-ask --interactive

Backend Selection¶

# Auto-detect best available backend
ansai-ask "your question"

# Use specific backend
ansai-ask --backend ollama "your question"
ansai-ask --backend groq "your question"

Webhook Alerts (Slack, Discord)¶

Instead of email, send alerts to Slack, Discord, or any webhook.

Configuration¶

# In your playbook or defaults/main.yml
alert_method: webhook      # email, webhook, both, none
webhook_url: "https://hooks.slack.com/services/xxx/yyy/zzz"
webhook_format: slack      # slack, discord, generic

Supported Formats¶

Slack - Rich formatted messages with attachments
Discord - Embed messages with colors and fields
Generic - JSON payload for custom integrations

What's Next¶

We're actively building more AI capabilities:

ansai-ask - Chat with AI about your infrastructure ✅
Ollama support - Local AI, no API keys ✅
Webhook alerts - Slack, Discord integration ✅
Web Dashboard - Visual AI insights
Predictive Alerts - AI predicts failures before they happen
Auto-tuning - AI optimizes your service configurations

Have ideas? Open a discussion or contribute!