#!/bin/bash
# Train AGI on CodeFix Marketing & Revenue Generation
# Uses Claude API to generate high-quality training data
# Budget: Up to 500 API credits

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
AGI_HOME="$(dirname "$SCRIPT_DIR")"
TRAINING_DATA="$AGI_HOME/training/codefix_marketing_training.txt"
API_KEY="${API_KEY:[credential removed]}"
API_ENDPOINT="https://api.anthropic.com/v1/messages"
API_MODEL="claude-3-5-sonnet-20241022"

TRAINING_LOG="$AGI_HOME/logs/codefix_training.log"
KNOWLEDGE_DB="$AGI_HOME/core/knowledge/codefix_knowledge.txt"
CREDITS_USED=0
MAX_CREDITS=500

echo "╔════════════════════════════════════════════════════════╗"
echo "║   🧠 CodeFix Marketing & Revenue Generation Training  ║"
echo "╚════════════════════════════════════════════════════════╝"
echo ""
echo "Training Data: $TRAINING_DATA"
echo "API Model: $API_MODEL"
echo "Budget: $MAX_CREDITS credits"
echo ""

# Create directories
mkdir -p "$AGI_HOME/logs"
mkdir -p "$AGI_HOME/core/knowledge"

# Function to call Claude API
call_api() {
    local prompt="$1"
    local response_file="$2"

    local request_json=$(cat <<EOF
{
  "model": "$API_MODEL",
  "max_tokens": 4096,
  "messages": [{
    "role": "user",
    "content": "$prompt"
  }]
}
EOF
)

    local http_code=$(curl -s -w "%{http_code}" -o "$response_file" \
        -X POST "$API_ENDPOINT" \
        -H "x-api-key: $API_KEY" \
        -H "content-type: application/json" \
        -d "$request_json")

    echo "$http_code"
}

# Extract Q&A pairs from training data
extract_qa_pairs() {
    grep -E "^Q:|^A:" "$TRAINING_DATA" | \
    awk 'BEGIN {q=""; a=""}
         /^Q:/ {if (q!="" && a!="") print q "|" a; q=$0; a=""}
         /^A:/ {a=$0}
         END {if (q!="" && a!="") print q "|" a}'
}

echo "📊 Extracting Q&A pairs from training data..."
QA_PAIRS=$(extract_qa_pairs)
QA_COUNT=$(echo "$QA_PAIRS" | wc -l)
echo "   Found $QA_COUNT Q&A pairs"
echo ""

echo "🤖 Training AGI using Claude API..."
echo "   This will generate high-quality responses and store them in the knowledge base"
echo ""

# Initialize knowledge base
cat > "$KNOWLEDGE_DB" <<EOF
# CodeFix Marketing & Revenue Generation Knowledge Base
# Generated by Claude API training
# $(date)

EOF

TRAINED_COUNT=0
BATCH_SIZE=10

while IFS='|' read -r question answer; do
    TRAINED_COUNT=$((TRAINED_COUNT + 1))

    # Progress indicator
    if [ $((TRAINED_COUNT % BATCH_SIZE)) -eq 0 ]; then
        PROGRESS=$((TRAINED_COUNT * 100 / QA_COUNT))
        echo "   Progress: $TRAINED_COUNT/$QA_COUNT ($PROGRESS%) - Credits used: ~$CREDITS_USED"
    fi

    # Clean up question and answer
    Q_CLEAN=$(echo "$question" | sed 's/^Q: //')
    A_CLEAN=$(echo "$answer" | sed 's/^A: //')

    # For every 5th question, enhance with Claude API
    if [ $((TRAINED_COUNT % 5)) -eq 0 ] && [ $CREDITS_USED -lt $MAX_CREDITS ]; then
        # Create enhanced prompt
        ENHANCE_PROMPT="You are a CodeFix marketing and sales expert. Given this Q&A about CodeFix, provide 3 additional insights or examples that would help someone better understand and sell this capability:\n\nQ: $Q_CLEAN\nA: $A_CLEAN\n\nProvide 3 concrete, actionable insights:"

        # Call API
        RESPONSE_FILE="/tmp/claude_response_$TRAINED_COUNT.json"
        HTTP_CODE=$(call_api "$ENHANCE_PROMPT" "$RESPONSE_FILE")

        if [ "$HTTP_CODE" = "200" ]; then
            # Extract response content
            ENHANCED=$(cat "$RESPONSE_FILE" | grep -o '"text":"[^"]*"' | head -1 | sed 's/"text":"//;s/"$//' | sed 's/\\n/\n/g')

            # Add to knowledge base
            cat >> "$KNOWLEDGE_DB" <<EOF

## $Q_CLEAN

**Answer**: $A_CLEAN

**Enhanced Insights**:
$ENHANCED

---

EOF
            CREDITS_USED=$((CREDITS_USED + 10))  # Estimate 10 credits per API call
        else
            # API failed, just use basic answer
            cat >> "$KNOWLEDGE_DB" <<EOF

## $Q_CLEAN

**Answer**: $A_CLEAN

---

EOF
        fi
    else
        # No API call, just store the Q&A
        cat >> "$KNOWLEDGE_DB" <<EOF

## $Q_CLEAN

**Answer**: $A_CLEAN

---

EOF
    fi

    # Stop if we've used too many credits
    if [ $CREDITS_USED -ge $MAX_CREDITS ]; then
        echo "   ⚠️  Reached credit limit ($MAX_CREDITS). Stopping API calls."
        echo "   Remaining Q&A pairs will be added without enhancement."
        break
    fi

done <<< "$QA_PAIRS"

echo ""
echo "✅ Training complete!"
echo "   Total Q&A pairs processed: $TRAINED_COUNT"
echo "   API credits used: ~$CREDITS_USED"
echo "   Knowledge base: $KNOWLEDGE_DB"
echo ""

# Now wake up the AGI agents and load the knowledge
echo "🚀 Waking up AGI agents..."
echo ""

# Start gossip service if not running
if ! curl -s -m 2 http://localhost:8008/stats > /dev/null 2>&1; then
    echo "   Starting Gossip Service..."
    cd "$AGI_HOME/gossip"
    if [ -f "gossipd" ]; then
        nohup ./gossipd > "$AGI_HOME/logs/gossip.log" 2>&1 &
        sleep 2
        echo "   ✅ Gossip Service started"
    else
        echo "   ⚠️  Gossip binary not found. Building..."
        go build -o gossipd main.go
        nohup ./gossipd > "$AGI_HOME/logs/gossip.log" 2>&1 &
        sleep 2
        echo "   ✅ Gossip Service built and started"
    fi
else
    echo "   ✅ Gossip Service already running"
fi

# Check agent count
AGENT_STATS=$(curl -s http://localhost:8008/stats 2>/dev/null || echo "error")
if [ "$AGENT_STATS" != "error" ]; then
    AGENT_COUNT=$(echo "$AGENT_STATS" | grep -o '"agent_count":[0-9]*' | cut -d: -f2)
    echo "   📊 Current agents: $AGENT_COUNT"

    # If agent count is low, spawn more
    if [ "$AGENT_COUNT" -lt 1000 ]; then
        echo "   ⚡ Spawning more agents to reach 10,000..."
        curl -s -X POST http://localhost:8008/spawn -d '{"count":10000}' > /dev/null 2>&1
        sleep 1
        AGENT_STATS=$(curl -s http://localhost:8008/stats 2>/dev/null)
        NEW_COUNT=$(echo "$AGENT_STATS" | grep -o '"agent_count":[0-9]*' | cut -d: -f2)
        echo "   ✅ Agent count increased to: $NEW_COUNT"
    fi
else
    echo "   ⚠️  Cannot connect to Gossip Service"
fi

echo ""
echo "╔════════════════════════════════════════════════════════╗"
echo "║   🎉 CodeFix Marketing Training Complete!             ║"
echo "╚════════════════════════════════════════════════════════╝"
echo ""
echo "Knowledge Base Stats:"
echo "  → Q&A Pairs: $TRAINED_COUNT"
echo "  → API Credits: ~$CREDITS_USED / $MAX_CREDITS"
echo "  → Knowledge DB: $KNOWLEDGE_DB ($(wc -l < "$KNOWLEDGE_DB") lines)"
echo ""
echo "AGI Status:"
echo "  → Gossip Service: http://localhost:8008/stats"
echo "  → Agents Active: $(echo "$AGENT_STATS" | grep -o '"agent_count":[0-9]*' | cut -d: -f2 || echo "N/A")"
echo "  → Training Log: $TRAINING_LOG"
echo ""
echo "Next Steps:"
echo "  1. Test AGI knowledge: curl http://localhost:8008/query -d '{\"q\":\"What is CodeFix?\"}'"
echo "  2. View agent stats: curl http://localhost:8008/stats"
echo "  3. Check training log: tail -f $TRAINING_LOG"
echo ""
echo "🚀 The AGI is now trained and ready to assist with CodeFix marketing!"
echo ""