src/core/ai.py

"""AI integration for mission analysis."""
import os
import json
import re
import aiohttp
from typing import Dict, List, Any, Optional

from .knowledge_loader import KnowledgeBaseLoader
from .ai_personality import AIPersonality, ConversationState, UserExperience
from .geocoding import GeocodingService


class Gemini2FlashAI:
    """Gemini 2.0 Flash AI integration with fallback."""
    
    def __init__(self):
        api_keys_str = os.getenv('GEMINI_API_KEYS', '')
        self.api_keys = [key.strip() for key in api_keys_str.split(',') if key.strip()]
        self.enabled = bool(self.api_keys)
        self.api_endpoint = "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-exp:generateContent"
        self.current_key_index = 0

        # Enhanced API key rotation system
        self.key_status = {}  # Track status of each key
        self.key_last_used = {}  # Track last usage time for each key
        self.key_cooldown_until = {}  # Track cooldown periods for rate limited keys
        self.initialize_key_tracking()
        
        # Load knowledge base for enhanced context
        self.knowledge_loader = KnowledgeBaseLoader()
        self.knowledge_base = self.knowledge_loader.load_complete_knowledge_base()
        
        # Initialize AI personality for user guidance
        self.personality = AIPersonality()
        
        # Initialize geocoding service (reads GOONG_API_KEY from env)
        self.geocoding = GeocodingService()

        # Note: chat_history will be provided by mission_planner context
        
    def initialize_key_tracking(self):
        """Initialize tracking for all API keys."""
        from datetime import datetime
        current_time = datetime.now()

        for i, key in enumerate(self.api_keys):
            self.key_status[key] = 'available'  # available, rate_limited, cooldown
            self.key_last_used[key] = current_time
            self.key_cooldown_until[key] = None

    def get_next_available_key(self) -> str:
        """Get the next available API key with intelligent rotation."""
        if not self.api_keys:
            return ""

        from datetime import datetime, timedelta
        current_time = datetime.now()

        # Try each key to find one that's available
        for attempt in range(len(self.api_keys)):
            candidate_key = self.api_keys[self.current_key_index]
            self.current_key_index = (self.current_key_index + 1) % len(self.api_keys)

            # Check if key is available
            if self.key_status[candidate_key] == 'available':
                self.key_last_used[candidate_key] = current_time
                return candidate_key

            # Check if rate limited key is ready for retry (after cooldown)
            elif self.key_status[candidate_key] == 'rate_limited':
                if self.key_cooldown_until[candidate_key] and current_time >= self.key_cooldown_until[candidate_key]:
                    print(f"🔄 API key {candidate_key[:20]}... cooldown period ended, retrying")
                    self.key_status[candidate_key] = 'available'
                    self.key_cooldown_until[candidate_key] = None
                    self.key_last_used[candidate_key] = current_time
                    return candidate_key

            # Check if key is in cooldown
            elif self.key_status[candidate_key] == 'cooldown':
                if self.key_cooldown_until[candidate_key] and current_time >= self.key_cooldown_until[candidate_key]:
                    print(f"🔄 API key {candidate_key[:20]}... cooldown period ended, making available")
                    self.key_status[candidate_key] = 'available'
                    self.key_last_used[candidate_key] = current_time
                    return candidate_key

        # If all keys are unavailable, return the first one anyway (better than no response)
        print(f"⚠️ All API keys are rate limited, using first key as fallback")
        return self.api_keys[0] if self.api_keys else ""

    def mark_key_rate_limited(self, api_key: str):
        """Mark an API key as rate limited and set cooldown."""
        from datetime import datetime, timedelta

        if api_key in self.key_status:
            self.key_status[api_key] = 'rate_limited'
            # Set cooldown for 60 seconds (adjust as needed based on API limits)
            self.key_cooldown_until[api_key] = datetime.now() + timedelta(seconds=60)
            print(f"🚫 API key {api_key[:20]}... marked as rate limited, cooldown until {self.key_cooldown_until[api_key]}")

    def mark_key_error(self, api_key: str):
        """Mark an API key as having an error and set shorter cooldown."""
        from datetime import datetime, timedelta

        if api_key in self.key_status:
            self.key_status[api_key] = 'cooldown'
            # Set shorter cooldown for general errors (30 seconds)
            self.key_cooldown_until[api_key] = datetime.now() + timedelta(seconds=30)
            print(f"⚠️ API key {api_key[:20]}... marked for cooldown until {self.key_cooldown_until[api_key]}")

    def mark_key_success(self, api_key: str):
        """Mark an API key as successful and available."""
        if api_key in self.key_status:
            self.key_status[api_key] = 'available'
            self.key_cooldown_until[api_key] = None
            print(f"✅ API key {api_key[:20]}... marked as available after successful use")

    def get_current_api_key(self) -> str:
        """Get API key with intelligent rotation and fallback."""
        return self.get_next_available_key()
        
    async def analyze_conversation(self, user_message: str, chat_history: List[Dict]) -> Dict[str, Any]:
        """🧠 GEMINI-FIRST CONVERSATION ANALYSIS: Intelligent, personalized responses."""
        
        # Always use Gemini for conversation analysis if available
        if self.enabled:
            return await self._gemini_conversation_analysis(user_message, chat_history)
        else:
            # Fallback to local personality system
            return await self._local_conversation_analysis(user_message, chat_history)
    
    async def _gemini_conversation_analysis(self, user_message: str, chat_history: List[Dict]) -> Dict[str, Any]:
        """🧠 Gemini-powered conversation analysis for intelligent responses."""
        
        # Determine user experience and conversation state using local tools
        user_experience = self.personality.analyze_user_experience(user_message, chat_history)
        conversation_state = self.personality.determine_conversation_state(user_message, chat_history)

        # Check if this is a mission request or needs guidance
        if conversation_state in [ConversationState.READY_TO_GENERATE]:
            return await self.analyze_mission_request(user_message)
        elif conversation_state == ConversationState.DRONE_STATUS_REQUEST:
            # Handle drone status requests
            return await self._handle_drone_status_request()
        elif conversation_state == ConversationState.GENERAL_QUERY:
            # Use Gemini for intelligent general query responses
            return await self._gemini_general_response(user_message, chat_history, user_experience)
        else:
            # Use Gemini for intelligent guidance responses  
            return await self._gemini_guidance_response(user_message, chat_history, conversation_state, user_experience)
    
    async def _local_conversation_analysis(self, user_message: str, chat_history: List[Dict]) -> Dict[str, Any]:
        """Local fallback conversation analysis."""
        # Determine user experience and conversation state
        user_experience = self.personality.analyze_user_experience(user_message, chat_history)
        conversation_state = self.personality.determine_conversation_state(user_message, chat_history)
        
        # Check if this is a mission request or needs guidance
        if conversation_state in [ConversationState.READY_TO_GENERATE]:
            return await self.analyze_mission_request(user_message)
        elif conversation_state == ConversationState.DRONE_STATUS_REQUEST:
            # Handle drone status requests
            return await self._handle_drone_status_request()
        elif conversation_state == ConversationState.GENERAL_QUERY:
            # Handle general queries
            guidance_response = self.personality.get_guidance_response(
                conversation_state, user_experience, {}, chat_history
            )
            return {
                'is_guidance': True,
                'conversation_state': conversation_state.value,
                'user_experience': user_experience.value,
                'guidance_response': guidance_response,
                'needs_mission_generation': False,
                'is_general_query': True
            }
        else:
            # Provide user guidance
            context = {
                'location': self._extract_location_from_history(chat_history),
                'mission_type': self._extract_mission_type_from_history(chat_history)
            }
            
            guidance_response = self.personality.get_guidance_response(
                conversation_state, user_experience, context, chat_history
            )
            
            return {
                'is_guidance': True,
                'conversation_state': conversation_state.value,
                'user_experience': user_experience.value,
                'guidance_response': guidance_response,
                'needs_mission_generation': False
            }
    
    async def _gemini_general_response(self, user_message: str, chat_history: List[Dict], user_experience) -> Dict[str, Any]:
        """🧠 Gemini-powered general query responses with flight awareness."""
        
        # Check for active flights to provide flight-aware responses
        flight_context = await self._get_flight_context()
        
        # Prepare chat context
        recent_context = ""
        if chat_history:
            recent_messages = chat_history[-3:]
            context_parts = []
            for msg in recent_messages:
                if 'user' in msg:
                    context_parts.append(f"User: {msg['user']}")
                if 'ai' in msg:
                    context_parts.append(f"Assistant: {msg['ai'][:100]}...")
            recent_context = "\n".join(context_parts)
        
        prompt = f"""You are DroneBot, a friendly and intelligent AI assistant for drone mission planning.
The user is asking a general question or greeting. Respond naturally and helpfully.

USER EXPERIENCE LEVEL: {user_experience.value}
USER MESSAGE: "{user_message}"

RECENT CHAT CONTEXT:
{recent_context}

{flight_context}

Provide a friendly, personalized response that:
- Shows you understand their message
- If there are active flights, proactively offer flight management options
- Maintains a conversational, friendly tone
- Offers helpful guidance about drone missions
- During active flights, suggest options like: "Do you want to check flight status?", "Pause the current mission?", "Update the flight plan?", "Have the drone return home?"
- For greetings, include current drone status if available

Keep the response conversational and engaging, around 100-200 words."""

        try:
            api_key = self.get_current_api_key()
            headers = {'Content-Type': 'application/json', 'x-goog-api-key': api_key}
            
            payload = {
                "contents": [{"parts": [{"text": prompt}]}],
                "generationConfig": {
                    "temperature": 0.7,  # More creative for conversation
                    "topK": 40,
                    "topP": 0.95,
                    "maxOutputTokens": 1024
                }
            }
            
            async with aiohttp.ClientSession() as session:
                async with session.post(self.api_endpoint, json=payload, headers=headers) as response:
                    if response.status == 200:
                        result = await response.json()
                        gemini_response = result['candidates'][0]['content']['parts'][0]['text']
                        
                        return {
                            'is_guidance': True,
                            'conversation_state': 'general_query',
                            'user_experience': user_experience.value,
                            'guidance_response': gemini_response,
                            'needs_mission_generation': False,
                            'is_general_query': True,
                            'gemini_powered': True
                        }
        except Exception as e:
            print(f"🚨 Gemini general response failed: {e}")
        
        # Fallback to local response
        guidance_response = self.personality.get_guidance_response(
            ConversationState.GENERAL_QUERY, user_experience, {}, chat_history
        )
        return {
            'is_guidance': True,
            'conversation_state': 'general_query',
            'user_experience': user_experience.value,
            'guidance_response': guidance_response,
            'needs_mission_generation': False,
            'is_general_query': True
        }
    
    async def _gemini_guidance_response(self, user_message: str, chat_history: List[Dict], 
                                      conversation_state, user_experience) -> Dict[str, Any]:
        """🧠 Gemini-powered guidance responses."""
        
        # Extract context
        location = self._extract_location_from_history(chat_history)
        mission_type = self._extract_mission_type_from_history(chat_history)
        
        # Prepare chat context
        recent_context = ""
        if chat_history:
            recent_messages = chat_history[-2:]
            context_parts = []
            for msg in recent_messages:
                if 'user' in msg:
                    context_parts.append(f"User: {msg['user']}")
            recent_context = "\n".join(context_parts)
        
        prompt = f"""You are DroneBot, a friendly and intelligent AI assistant for drone mission planning.
The user needs guidance to complete their drone mission request.

USER MESSAGE: "{user_message}"
CONVERSATION STATE: {conversation_state.value}
USER EXPERIENCE: {user_experience.value}
KNOWN LOCATION: {location}
KNOWN MISSION TYPE: {mission_type}

RECENT CONTEXT:
{recent_context}

Based on the conversation state, provide helpful, personalized guidance:
- If they need location: Ask for coordinates or location description with examples
- If they need mission type: Explain mission types and ask what they want to accomplish
- If they need details: Ask for specific requirements like altitude, area size, etc.
- Be encouraging and educational for beginners
- Be concise and technical for experts

Provide a friendly, helpful response that guides them to the next step."""

        try:
            api_key = self.get_current_api_key()
            headers = {'Content-Type': 'application/json', 'x-goog-api-key': api_key}
            
            payload = {
                "contents": [{"parts": [{"text": prompt}]}],
                "generationConfig": {
                    "temperature": 0.6,
                    "topK": 40,
                    "topP": 0.95,
                    "maxOutputTokens": 1024
                }
            }
            
            async with aiohttp.ClientSession() as session:
                async with session.post(self.api_endpoint, json=payload, headers=headers) as response:
                    if response.status == 200:
                        result = await response.json()
                        gemini_response = result['candidates'][0]['content']['parts'][0]['text']
                        
                        return {
                            'is_guidance': True,
                            'conversation_state': conversation_state.value,
                            'user_experience': user_experience.value,
                            'guidance_response': gemini_response,
                            'needs_mission_generation': False,
                            'gemini_powered': True
                        }
        except Exception as e:
            print(f"🚨 Gemini guidance response failed: {e}")
        
        # Fallback to local response
        context = {'location': location, 'mission_type': mission_type}
        guidance_response = self.personality.get_guidance_response(
            conversation_state, user_experience, context, chat_history
        )
        return {
            'is_guidance': True,
            'conversation_state': conversation_state.value,
            'user_experience': user_experience.value,
            'guidance_response': guidance_response,
            'needs_mission_generation': False
        }

    async def _handle_drone_status_request(self) -> Dict[str, Any]:
        """Handle drone status requests by calling the drone status endpoint."""
        try:
            from ..core.drone_status import drone_status

            current_status = drone_status.get_current_status()
            constraints = drone_status.get_flight_constraints()

            # Format the drone status response
            status_message = self._format_drone_status_response(current_status, constraints)

            return {
                'is_guidance': False,
                'is_drone_status': True,
                'conversation_state': 'drone_status_request',
                'user_experience': 'unknown',
                'drone_status_response': status_message,
                'needs_mission_generation': False,
                'drone_status': current_status,
                'flight_constraints': constraints
            }
        except Exception as e:
            return {
                'is_guidance': False,
                'is_drone_status': True,
                'conversation_state': 'drone_status_request',
                'error': f"Error retrieving drone status: {str(e)}",
                'drone_status_response': f"❌ Error retrieving drone status: {str(e)}",
                'needs_mission_generation': False
            }

    def _format_drone_status_response(self, current_status: Dict, constraints: Dict) -> str:
        """Format drone status information into a user-friendly message."""
        message_parts = []

        message_parts.append("🚁 **Drone Status Report**")
        message_parts.append("")

        # Basic status information
        battery = current_status.get('batteryRemaining', 0)
        satellites = current_status.get('satellite', 0)
        voltage = current_status.get('voltage', 0.0)

        message_parts.append(f"🔋 **Battery:** {battery}% ({voltage:.1f}V)")
        message_parts.append(f"📡 **GPS:** {satellites} satellites")
        message_parts.append(f"🛩️ **Mode:** {current_status.get('mode', 'UNKNOWN')}")
        message_parts.append(f"📍 **Location:** {current_status.get('latitude', 0):.4f}, {current_status.get('longitude', 0):.4f}")
        message_parts.append(f"🛩️ **Altitude:** {current_status.get('altitude', 0):.0f}m")
        message_parts.append("")

        # Flight constraints
        message_parts.append("**Flight Constraints:**")
        message_parts.append(f"• Safe Flight Time: {constraints.get('safe_flight_time_minutes', 0)} minutes")
        message_parts.append(f"• Max Distance: {constraints.get('max_safe_distance_meters', 0)}m")
        message_parts.append(f"• Optimal Altitude: {constraints.get('optimal_altitude_range', (30, 80))[0]}-{constraints.get('optimal_altitude_range', (30, 80))[1]}m")
        message_parts.append(f"• Ready to Fly: {'✅ Yes' if constraints.get('ready_to_fly', False) else '❌ No'}")
        message_parts.append("")

        if not constraints.get('ready_to_fly', False):
            message_parts.append("⚠️ **Not ready to fly:** Check battery level, GPS connection, and safety constraints.")

        message_parts.append("")
        message_parts.append("For mission planning, the system will optimize altitude and safety parameters based on this status.")

        # Include raw telemetry hint for full status requests
        if current_status.get('data_source') == 'hardware_telemetry':
            message_parts.append("")
            message_parts.append("Raw telemetry available via /drone/status")

        return "\n".join(message_parts)
    
    async def analyze_mission_request(self, user_message: str) -> Dict[str, Any]:
        """🧠 GEMINI-FIRST ANALYSIS: Always use Gemini as primary AI brain with tool support."""
        
        # ALWAYS use Gemini if available - it's the smart brain that coordinates everything
        if self.enabled:
            print(f"🧠 GEMINI-FIRST: Using Gemini 2.0 Flash as primary AI brain")
            return await self._gemini_first_analysis(user_message)
        else:
            print(f"⚠️ Gemini not available, using local fallback. API keys: {len(self.api_keys)}")
            return await self._fallback_analysis(user_message, [])
    
    async def _gemini_first_analysis(self, user_message: str) -> Dict[str, Any]:
        """🧠 Gemini-first analysis with intelligent tool integration."""

        print(f"🔍 DEBUG: Starting Gemini analysis for: {user_message[:50]}...")

        # Pre-extract supporting data for Gemini
        explicit_coords = self._extract_coordinates(user_message)
        is_location_query = self._is_location_based_query(user_message)

        print(f"🔍 DEBUG: Explicit coords: {explicit_coords}, Is location query: {is_location_query}")

        # Prepare comprehensive context for Gemini
        support_data = {
            "explicit_coordinates": explicit_coords,
            "has_location_query": is_location_query,
            "geocoding_available": True,
            "drone_status_available": True
        }

        print(f"🔍 DEBUG: Support data prepared: {support_data}")

        # Add geocoding if needed
        geocoding_info = ""
        if is_location_query and not explicit_coords:
            print(f"🔍 DEBUG: Attempting geocoding for location query")
            try:
                geocoding_result = await self.geocoding.parse_location_with_distance(user_message)
                if geocoding_result:
                    support_data["geocoded_location"] = geocoding_result
                    geocoding_info = f"""
GEOCODING RESULT:
- Location: {geocoding_result['location_name']}
- Coordinates: {geocoding_result['coordinates']}
- Distance: {geocoding_result.get('distance_meters', 'N/A')}m"""
                    print(f"🔍 DEBUG: Geocoding successful: {geocoding_result}")
                else:
                    print(f"❌ DEBUG: Geocoding failed - no result")
            except Exception as e:
                geocoding_info = f"Geocoding failed: {e}"
                print(f"❌ DEBUG: Geocoding exception: {e}")
        else:
            print(f"🔍 DEBUG: Skipping geocoding - has explicit coords or not location query")
        
        # Enhanced prompt for Gemini with professional AI-driven analysis
        prompt = f"""You are DroneBot, a professional AI drone mission specialist with comprehensive knowledge of aviation, photogrammetry, and mission planning. You must create intelligent, optimized missions based on real-world requirements.

COMPREHENSIVE KNOWLEDGE BASE:
{self.knowledge_base}

USER REQUEST: "{user_message}"

ENVIRONMENTAL DATA:
- Coordinates: {explicit_coords}
- Location query: {is_location_query}
{geocoding_info}

PROFESSIONAL MISSION ANALYSIS REQUIREMENTS:
You must analyze the request using professional drone operations knowledge:

1. MISSION TYPE INTELLIGENCE:
   - Survey: Photogrammetry requirements, GSD calculation, overlap optimization
   - Patrol: Security patterns, visibility requirements, threat assessment
   - Photography: Lighting analysis, composition angles, equipment optimization
   - Inspection: Structure analysis, safety protocols, detailed examination patterns
   - Go_straight: Efficiency optimization, obstacle avoidance, direct routing

2. ENVIRONMENTAL ANALYSIS:
   - Calculate optimal altitude based on mission requirements and safety
   - Analyze terrain complexity and obstacle clearance needs
   - Consider weather impact on flight parameters
   - Assess battery requirements for mission completion

3. PROFESSIONAL FLIGHT PLANNING:
   - Generate waypoint patterns based on industry standards
   - Optimize for data quality, safety, and efficiency
   - Calculate proper overlap for survey missions (80% forward, 60% side)
   - Design approach angles for inspection missions
   - Plan hold times based on equipment requirements

4. SAFETY INTEGRATION:
   - Assess flight risks and mitigation strategies
   - Calculate battery reserves and return-to-home requirements
   - Ensure regulatory compliance and airspace considerations
   - Plan emergency procedures and abort criteria

As a professional AI mission planner, provide a comprehensive JSON response:

{{
  "mission_type": "survey|patrol|photography|inspection|go_straight|simple_flight",
  "confidence": 0.95,
  "coordinates": [latitude_6_decimal, longitude_6_decimal],
  "altitude": optimized_altitude_meters,
  "direction": "north|south|east|west|northeast|northwest|southeast|southwest|null",
  "distance": calculated_distance_meters,
  "reasoning": "Professional analysis explaining mission optimization, safety considerations, and technical decisions",
  "mission_parameters": {{
    "ground_sample_distance": gsd_cm_per_pixel,
    "forward_overlap": overlap_percentage,
    "side_overlap": overlap_percentage,
    "flight_speed": optimized_speed_ms,
    "hold_time": required_hold_time_seconds,
    "camera_trigger_distance": trigger_distance_meters,
    "battery_consumption_estimate": estimated_wh_usage,
    "flight_time_estimate": estimated_minutes,
    "data_volume_estimate": estimated_gb,
    "weather_considerations": ["wind_impact", "visibility_requirements"],
    "equipment_requirements": ["camera_specs", "gimbal_settings"]
  }},
  "waypoint_strategy": "Detailed explanation of flight pattern optimization based on professional standards",
  "safety_considerations": [
    "Comprehensive safety analysis",
    "Risk mitigation strategies", 
    "Emergency procedures",
    "Regulatory compliance checks"
  ],
  "professional_insights": {{
    "mission_complexity": "low|medium|high",
    "skill_level_required": "beginner|intermediate|expert",
    "environmental_challenges": ["challenge1", "challenge2"],
    "optimization_opportunities": ["improvement1", "improvement2"],
    "quality_assurance": "Data quality predictions and recommendations"
  }},
  "personality_response": "Friendly, professional message explaining the mission plan with educational insights"
}}

IMPORTANT:
- Be friendly, understanding, and personalized in your reasoning
- Use the support tools data (coordinates, geocoding) intelligently
- If coordinates are missing, include helpful guidance in personality_response
- Provide detailed, human-like reasoning that shows you understand the user's intent
- Make safety recommendations based on mission type

Respond ONLY with valid JSON."""

        try:
            api_key = self.get_current_api_key()
            if not api_key:
                print("❌ No API key available for Gemini")
                return await self._fallback_analysis(user_message, [])

            print(f"🔍 DEBUG: Using API key: {api_key[:20]}...")
            headers = {'Content-Type': 'application/json', 'x-goog-api-key': api_key}
            
            payload = {
                "contents": [{"parts": [{"text": prompt}]}],
                "generationConfig": {
                    "temperature": 0.4,  # Slightly more creative for personality
                    "topK": 40,
                    "topP": 0.95,
                    "maxOutputTokens": 3072
                }
            }
            
            print(f"🔍 DEBUG: Making Gemini API call...")
            async with aiohttp.ClientSession() as session:
                async with session.post(self.api_endpoint, json=payload, headers=headers) as response:
                    print(f"🔍 DEBUG: Gemini API response status: {response.status}")

                    # Handle rate limiting by marking key and trying next available key
                    if response.status == 429:
                        print(f"🚫 API key {api_key[:20]}... got rate limited (429)")
                        self.mark_key_rate_limited(api_key)

                        # Try with next available key before falling back to local AI
                        next_key = self.get_next_available_key()
                        if next_key and next_key != api_key:
                            print(f"🔄 Trying with next available key: {next_key[:20]}...")
                            retry_headers = headers.copy()
                            retry_headers['x-goog-api-key'] = next_key
                            async with aiohttp.ClientSession() as retry_session:
                                async with retry_session.post(self.api_endpoint, json=payload, headers=retry_headers) as retry_response:
                                    print(f"🔄 Retry response status: {retry_response.status}")
                                    if retry_response.status == 200:
                                        result = await retry_response.json()
                                        print(f"🔄 Retry successful! Processing with new key")
                                        # Continue with the successful result and process it
                                        print(f"🔄 Retry result keys: {list(result.keys()) if isinstance(result, dict) else 'not dict'}")

                                        # Process the successful retry result
                                        if not result or 'candidates' not in result or not result['candidates']:
                                            print("❌ Invalid retry response structure")
                                            return await self._fallback_analysis(user_message, [])

                                        candidate = result['candidates'][0]
                                        if 'content' not in candidate or 'parts' not in candidate['content'] or not candidate['content']['parts']:
                                            print("❌ Missing content in retry response")
                                            return await self._fallback_analysis(user_message, [])

                                        content = candidate['content']['parts'][0]['text']
                                        print(f"🔄 Retry content: {content[:200]}...")

                                        # Clean JSON
                                        content = content.strip()
                                        if content.startswith('```json'):
                                            content = content[7:-3]
                                        elif content.startswith('```'):
                                            content = content[3:-3]

                                        try:
                                            analysis = json.loads(content)
                                            if analysis is None:
                                                print("❌ Retry returned null JSON")
                                                return await self._fallback_analysis(user_message, [])
                                        except json.JSONDecodeError as e:
                                            print(f"❌ Invalid JSON from retry: {e}")
                                            return await self._fallback_analysis(user_message, [])

                                        # Enhance with support tool data
                                        if explicit_coords and not analysis.get('coordinates'):
                                            analysis['coordinates'] = explicit_coords
                                        elif 'geocoded_location' in support_data and not analysis.get('coordinates'):
                                            analysis['coordinates'] = list(support_data['geocoded_location']['coordinates'])

                                        analysis['gemini_enhanced'] = True
                                        analysis['model'] = 'gemini-2.0-flash'
                                        analysis['support_tools_used'] = list(support_data.keys())

                                        # Mark the successful key as available for future use
                                        self.mark_key_success(next_key)

                                        print(f"🔄 Retry analysis successful: {analysis}")
                                        return analysis
                                    elif retry_response.status == 429:
                                        print(f"🚫 Next key also rate limited, falling back to local AI")
                                        self.mark_key_rate_limited(next_key)
                                        return await self._fallback_analysis(user_message, [])
                                    else:
                                        print(f"❌ Retry failed with status {retry_response.status}, falling back to local AI")
                                        self.mark_key_error(next_key)
                                        return await self._fallback_analysis(user_message, [])
                        else:
                            print(f"⚠️ No other keys available, falling back to local AI")
                            return await self._fallback_analysis(user_message, [])
                    elif response.status == 200:
                        result = await response.json()
                        print(f"🔍 DEBUG: Gemini API response received, keys: {list(result.keys()) if isinstance(result, dict) else 'not dict'}")
                        print(f"🔍 DEBUG: Full result: {result}")

                        # Debug: Check if result structure is correct
                        if not result or 'candidates' not in result or not result['candidates']:
                            print("❌ Invalid Gemini API response structure")
                            return await self._fallback_analysis(user_message, [])

                        candidate = result['candidates'][0]
                        print(f"🔍 DEBUG: Candidate keys: {list(candidate.keys()) if isinstance(candidate, dict) else 'not dict'}")

                        if 'content' not in candidate or 'parts' not in candidate['content'] or not candidate['content']['parts']:
                            print("❌ Missing content in Gemini response")
                            return await self._fallback_analysis(user_message, [])

                        content = candidate['content']['parts'][0]['text']
                        print(f"🔍 DEBUG: Raw content from Gemini: {content[:200]}...")
                        
                        # Clean JSON
                        content = content.strip()
                        if content.startswith('```json'):
                            content = content[7:-3]
                        elif content.startswith('```'):
                            content = content[3:-3]
                            
                        print(f"🔍 DEBUG: Cleaned content: {content[:200]}...")

                        # Debug: Check if content is valid JSON
                        try:
                            analysis = json.loads(content)
                            print(f"🔍 DEBUG: Parsed analysis: {analysis}")
                            if analysis is None:
                                print("❌ Gemini returned null JSON")
                                return await self._fallback_analysis(user_message, [])
                        except json.JSONDecodeError as e:
                            print(f"❌ Invalid JSON from Gemini: {e}")
                            print(f"Content: {content[:200]}...")
                            return await self._fallback_analysis(user_message, [])

                        # Enhance with support tool data
                        if explicit_coords and not analysis.get('coordinates'):
                            analysis['coordinates'] = explicit_coords
                        elif 'geocoded_location' in support_data and not analysis.get('coordinates'):
                            analysis['coordinates'] = list(support_data['geocoded_location']['coordinates'])

                        analysis['gemini_enhanced'] = True
                        analysis['model'] = 'gemini-2.0-flash'
                        analysis['support_tools_used'] = list(support_data.keys())

                        # Mark the successful key as available for future use
                        self.mark_key_success(api_key)

                        print(f"🔍 DEBUG: Final analysis: {analysis}")
                        return analysis
                    elif response.status in [400, 401, 403]:
                        print(f"❌ Gemini API error ({response.status}), marking key for cooldown")
                        self.mark_key_error(api_key)
                        return await self._fallback_analysis(user_message, [])
                    else:
                        print(f"❌ Gemini API returned status {response.status}, marking key for cooldown")
                        self.mark_key_error(api_key)
                        return await self._fallback_analysis(user_message, [])
                        
        except Exception as e:
            print(f"🚨 Gemini API failed: {e}")
            return await self._fallback_analysis(user_message, [])
    
    async def _fallback_analysis(self, user_message: str, chat_history: List[Dict] = None) -> Dict[str, Any]:
        """Local AI fallback analysis with context awareness."""
        if chat_history is None:
            chat_history = []
            
        mission_type = self._extract_mission_type(user_message)
        coordinates = await self._get_coordinates_from_context(user_message, chat_history)
        
        return {
            'mission_type': mission_type,
            'confidence': 0.75,
            'coordinates': coordinates,
            'altitude': self._extract_altitude(user_message),
            'direction': self._extract_direction(user_message),
            'distance': self._extract_distance(user_message),
            'reasoning': f"Smart local AI: Detected '{mission_type}' mission with intelligent context analysis",
            'mission_parameters': self._get_mission_parameters(mission_type),
            'waypoint_strategy': self._get_waypoint_strategy(mission_type),
            'safety_considerations': self._get_safety_considerations(mission_type),
            'gemini_enhanced': False,
            'model': 'local-fallback'
        }
    
    def _extract_mission_type(self, message: str) -> str:
        """Extract mission type from message with smart intent recognition."""
        message_lower = message.lower()
        
        # Priority-based mission type detection to avoid keyword conflicts

        # Priority 1: Specific mission type keywords (highest priority)
        if any(word in message_lower for word in ['survey', 'map', 'mapping', 'grid', 'coverage', 'scan', 'scanning']):
            return 'survey'
        elif any(word in message_lower for word in ['patrol', 'perimeter', 'guard', 'security', 'monitor', 'watch', 'circle', 'secure']):
            return 'patrol'
        elif any(word in message_lower for word in ['photo', 'photography', 'picture', 'capture', 'take photos', 'aerial photos', 'aerial', 'shoot']):
            return 'photography'
        elif any(word in message_lower for word in ['inspect', 'inspection', 'examine', 'check', 'analyze', 'detail', 'structure']):
            return 'inspection'

        # Priority 2: Directional movement keywords (medium priority)
        elif any(word in message_lower for word in ['go straight', 'straight line', 'fly straight', 'move straight', 'straight flight', 'direct flight']):
            return 'go_straight'
        elif any(word in message_lower for word in ['go south', 'go north', 'go east', 'go west', 'fly to', 'move to']):
            return 'go_straight'
        
        # Priority 3: General context keywords (lowest priority)
        mission_keywords = {
            'survey': ['field', 'farm', 'land', 'area coverage'],
            'patrol': ['route', 'boundary', 'surveillance'],
            'photography': ['building', 'structure', 'landscape'],
            'inspection': ['power line', 'bridge', 'roof', 'infrastructure'],
            'go_straight': ['straight distance', 'linear path', 'direct route']
        }

        for mission_type, keywords in mission_keywords.items():
            if any(keyword in message_lower for keyword in keywords):
                return mission_type
        
        return 'simple_flight'
    
    def _extract_coordinates(self, message: str) -> Optional[List[float]]:
        """Extract coordinates from message with strict validation to prevent false positives."""
        # STRICT coordinate patterns - must be explicitly GPS coordinates
        patterns = [
            # With explicit coordinate keywords (most reliable)
            r'(?:coordinates?|coords)\s+([-+]?\d+\.?\d*)[,\s]+([-+]?\d+\.?\d*)',
            r'(?:location|position)\s+([-+]?\d+\.?\d*)[,\s]+([-+]?\d+\.?\d*)',
            r'at\s+(?:coordinates?|coords)?\s*([-+]?\d+\.?\d*)[,\s]+([-+]?\d+\.?\d*)',
            
            # Latitude/longitude explicitly mentioned
            r'lat[itude]*[:\s]+([-+]?\d+\.?\d*)[,\s]+lon[gitude]*[:\s]+([-+]?\d+\.?\d*)',
            
            # Decimal coordinates (must have decimal points for both numbers)
            r'(\d+\.\d+)[,\s]+(\d+\.\d+)(?!\s*[a-zA-Z])',  # Not followed by letters
            
            # Standard coordinate format with "at" keyword
            r'at\s+(\d{1,2}\.\d{3,})[,\s]+(\d{2,3}\.\d{3,})',
            
            # Simple pattern but ONLY for reasonable coordinate ranges
            r'(?:^|\s|at\s+)((?:[1-9]\d?|1[0-7]\d)\.?\d*)[,\s]+((?:[1-9]\d{1,2}|[1-2]\d{2})\.?\d*)(?=\s|$)',
        ]

        for pattern in patterns:
            matches = re.findall(pattern, message, re.IGNORECASE)
            if matches:
                try:
                    # Handle different pattern formats
                    if len(matches[0]) == 2:
                        lat, lon = float(matches[0][0]), float(matches[0][1])
                    else:
                        # For patterns with more groups, take first two
                        lat, lon = float(matches[0][0]), float(matches[0][1])

                    # STRICT validation for GPS coordinates
                    if (-90 <= lat <= 90 and -180 <= lon <= 180 and
                        # Additional checks to prevent false positives
                        abs(lat) > 0.1 and abs(lon) > 0.1 and  # Not too close to 0,0
                        not (lat < 5 and lon < 5)):  # Avoid small numbers that might be measurements
                        return [lat, lon]
                except (ValueError, IndexError):
                    continue
        return None
    
    def _extract_altitude(self, message: str) -> int:
        """Extract altitude from message."""
        altitude_patterns = [r'(\d+)\s*m(?:eter)?s?\b', r'altitude\s+(\d+)']
        for pattern in altitude_patterns:
            matches = re.findall(pattern, message.lower())
            if matches:
                return int(matches[0])
        return 80
    
    def _extract_direction(self, message: str) -> str:
        """Extract direction from message."""
        directions = ['north', 'south', 'east', 'west', 'northeast', 'northwest', 'southeast', 'southwest']
        message_lower = message.lower()
        for direction in directions:
            if direction in message_lower:
                return direction
        return 'north'
    
    def _extract_distance(self, message: str) -> int:
        """Extract distance from message."""
        distance_patterns = [r'(\d+)\s*m(?:eter)?s?', r'(\d+)\s*km']
        for pattern in distance_patterns:
            matches = re.findall(pattern, message.lower())
            if matches:
                distance = int(matches[0])
                if 'km' in pattern:
                    distance *= 1000
                return distance
        return 100
    
    def _get_mission_parameters(self, mission_type: str) -> Dict[str, Any]:
        """Get mission-specific parameters."""
        parameters = {
            'survey': {'grid_size': 4, 'overlap': 0.7, 'spacing': 200},
            'patrol': {'perimeter_points': 8, 'radius': 300, 'hold_time': 2.0},
            'inspection': {'detail_points': 6, 'precision': 'high', 'slow_speed': True},
            'photography': {'photo_positions': 5, 'hold_time': 3.0, 'angles': 'multiple'},
            'go_straight': {'waypoints': 3, 'linear': True, 'direct_path': True},
            'simple_flight': {'waypoints': 4, 'pattern': 'basic', 'rectangular': True}
        }
        return parameters.get(mission_type, {})
    
    def _get_waypoint_strategy(self, mission_type: str) -> str:
        """Get waypoint strategy description."""
        strategies = {
            'survey': 'Grid pattern with optimal coverage and 70% overlap for mapping',
            'patrol': 'Perimeter pattern with 8 strategic points for surveillance',
            'inspection': 'Linear pattern with high precision waypoints for detailed examination',
            'photography': 'Strategic positioning with multiple angles and hold times',
            'go_straight': 'Direct linear path with evenly spaced waypoints',
            'simple_flight': 'Basic rectangular pattern for general flight operations'
        }
        return strategies.get(mission_type, 'Standard waypoint pattern optimized for mission type')
    
    def _get_safety_considerations(self, mission_type: str) -> List[str]:
        """Get safety considerations for mission type."""
        base_safety = [
            "Maintain 30% battery reserve minimum",
            "Check weather conditions before flight",
            "Ensure GPS lock before takeoff",
            "Verify no-fly zone compliance"
        ]
        
        specific_safety = {
            'survey': ["Maintain consistent altitude for mapping accuracy", "Monitor for aircraft in survey area"],
            'patrol': ["Be aware of security personnel", "Maintain communication during perimeter flight"],
            'inspection': ["Maintain safe distance from structures", "Use slow speeds for detailed observation"],
            'photography': ["Consider lighting conditions", "Plan for sun angle optimization"],
            'go_straight': ["Check for obstacles along flight path", "Ensure clear landing zone"],
            'simple_flight': ["Standard safety protocols apply", "Monitor for other aircraft"]
        }
        
        return base_safety + specific_safety.get(mission_type, [])
    
    def _extract_location_from_history(self, chat_history: List[Dict]) -> str:
        """Extract location information from chat history."""
        for message in reversed(chat_history[-5:]):  # Check last 5 messages
            user_msg = message.get('user', '')
            coords = self._extract_coordinates(user_msg)
            if coords:
                return f"{coords[0]}, {coords[1]}"
        return "not specified"
    
    async def _get_coordinates_from_context(self, message: str, chat_history: List[Dict]) -> Optional[List[float]]:
        """Get coordinates from current message, chat history, or geocoding service."""
        # First try current message
        coords = self._extract_coordinates(message)
        if coords:
            return coords
        
        # If this is a location-based query, prioritize geocoding over chat history
        if self._is_location_based_query(message):
            try:
                geocoded_coords = await self.geocoding.geocode_location(message)
                if geocoded_coords:
                    return list(geocoded_coords)
            except Exception as e:
                print(f"Geocoding failed: {e}")

        # Then try chat history as fallback
        for msg in reversed(chat_history[-3:]):
            user_msg = msg.get('user', '')
            coords = self._extract_coordinates(user_msg)
            if coords:
                return coords
        
        return None
    
    def _is_location_based_query(self, message: str) -> bool:
        """Check if the message is a location-based query that needs geocoding."""
        message_lower = message.lower()
        
        # SKIP geocoding if user provided explicit coordinates
        if self._extract_coordinates(message):
            return False

        # Check for location indicators that require geocoding
        location_indicators = [
            'around', 'near', 'in', 'district', 'province', 'city', 'town',
            'street', 'road', 'avenue', 'park', 'university', 'school',
            'hospital', 'airport', 'station', 'mall', 'market', 'center', 'from'
        ]

        has_location_words = any(f' {word} ' in f' {message_lower} ' for word in location_indicators)

        # Check for Vietnamese place name patterns
        vietnamese_places = ['đà nẵng', 'hải châu', 'sơn trà', 'ngũ hành sơn', 'liên chiểu', 'hội an', 'huế', 'quảng nam', 'doi cung', 'đội cung', 'hai chau']

        has_vietnamese_place = any(place in message_lower for place in vietnamese_places)

        # Check for distance patterns like "100m around"
        has_distance_pattern = bool(re.search(r'\d+\s*m\s+(around|from)', message_lower))

        # Must have either location words, Vietnamese places, or distance pattern
        # AND must have mission-related words like "scan", "survey", etc.
        mission_words = ['scan', 'survey', 'fly', 'mission', 'help me', 'go', 'straight']
        has_mission_intent = any(f' {word} ' in f' {message_lower} ' for word in mission_words)

        return (has_location_words or has_vietnamese_place or has_distance_pattern) and has_mission_intent
    
    def _extract_mission_type_from_history(self, chat_history: List[Dict]) -> str:
        """Extract mission type from chat history."""
        for message in reversed(chat_history[-5:]):
            user_msg = message.get('user', '')
            mission_type = self._extract_mission_type(user_msg)
            if mission_type != 'simple_flight':
                return mission_type
        return "not specified"
    
    async def _get_flight_context(self) -> str:
        """Get current flight context for AI responses."""
        try:
            from ..core.flight_manager import flight_manager
            
            # Get active missions
            active_missions = flight_manager.active_missions
            if not active_missions:
                return "FLIGHT STATUS: No active flights currently."
            
            flight_summaries = []
            for mission_id, mission in active_missions.items():
                # Get telemetry for this drone
                telemetry = flight_manager.drone_telemetry.get(mission.drone_id)
                
                completed_waypoints = sum(1 for wp in mission.waypoints if wp.completed)
                total_waypoints = len(mission.waypoints)
                progress = (completed_waypoints / total_waypoints) * 100 if total_waypoints > 0 else 0
                
                flight_info = f"""
ACTIVE FLIGHT:
- Mission: {mission_id} ({mission.state.value})
- Drone: {mission.drone_id}
- Progress: {completed_waypoints}/{total_waypoints} waypoints ({progress:.1f}%)"""
                
                if telemetry:
                    flight_info += f"""
- Battery: {telemetry.battery}%
- Location: {telemetry.lat:.4f}, {telemetry.lon:.4f}
- Altitude: {telemetry.alt:.1f}m"""
                
                flight_summaries.append(flight_info)
            
            context = "\n".join(flight_summaries)
            return f"CURRENT FLIGHT STATUS:\n{context}\n\nIMPORTANT: User may want to check status, pause, update, or cancel the active flight(s)."
            
        except Exception as e:
            return f"FLIGHT STATUS: Error retrieving flight info ({str(e)})"


class ChatHistory:
    """Enhanced chat history for conversational AI."""
    
    def __init__(self, max_messages: int = 10):
        self.messages = []
        self.max_messages = max_messages
        self.user_context = {
            'experience_level': None,
            'current_mission': None,
            'preferences': {},
            'last_coordinates': None
        }
        
    def add_message(self, user_message: str, ai_response: str, metadata: Dict = None):
        """Add message pair to history with metadata."""
        from datetime import datetime
        
        if metadata is None:
            metadata = {}
            
        self.messages.append({
            'user': user_message,
            'ai': ai_response,
            'timestamp': datetime.now().isoformat(),
            'metadata': metadata
        })
        
        # Update user context
        self._update_user_context(user_message, metadata)
        
        if len(self.messages) > self.max_messages:
            self.messages = self.messages[-self.max_messages:]
            
    def _update_user_context(self, user_message: str, metadata: Dict):
        """Update user context based on conversation."""
        message_lower = user_message.lower()
        
        # Update experience level
        if any(term in message_lower for term in ['new to drones', 'beginner', 'first time']):
            self.user_context['experience_level'] = 'beginner'
        elif any(term in message_lower for term in ['mavlink', 'qgc', 'frame', 'coordinate']):
            self.user_context['experience_level'] = 'expert'
        
        # Extract and store coordinates
        import re
        coord_pattern = r'([-+]?\d+\.?\d*)[,\s]+([-+]?\d+\.?\d*)'
        matches = re.findall(coord_pattern, user_message)
        if matches:
            try:
                lat, lon = float(matches[0][0]), float(matches[0][1])
                if -90 <= lat <= 90 and -180 <= lon <= 180:
                    self.user_context['last_coordinates'] = (lat, lon)
            except ValueError:
                pass
            
    def get_context(self) -> str:
        """Get intelligent conversation context."""
        if not self.messages:
            return "New conversation"
        
        context_parts = ["Recent conversation:"]
        for msg in self.messages[-3:]:
            context_parts.append(f"User: {msg['user']}")
            context_parts.append(f"AI: {msg['ai'][:100]}...")
            
        # Add user context
        if self.user_context['experience_level']:
            context_parts.append(f"User experience: {self.user_context['experience_level']}")
        if self.user_context['last_coordinates']:
            lat, lon = self.user_context['last_coordinates']
            context_parts.append(f"Last known coordinates: {lat:.6f}, {lon:.6f}")
            
        return "\n".join(context_parts)
    
    def get_user_experience(self) -> str:
        """Get user experience level."""
        return self.user_context.get('experience_level', 'unknown')
    
    def has_coordinates(self) -> bool:
        """Check if user has provided coordinates."""
        return self.user_context['last_coordinates'] is not None
    
    def get_last_coordinates(self) -> Optional[tuple]:
        """Get last known coordinates."""
        return self.user_context['last_coordinates']