AI-Life-Coach-Streamlit / core /coordinator.py
rdune71's picture
Implement manual HF endpoint activation with clear visual indications and smart research integration
ac83e06
import asyncio
import logging
from typing import List, Dict, Optional, AsyncGenerator
from core.llm_factory import llm_factory
from core.session import session_manager
from services.hf_endpoint_monitor import hf_monitor
from services.weather import weather_service
try:
from tavily import TavilyClient
TAVILY_AVAILABLE = True
except ImportError:
TavilyClient = None
TAVILY_AVAILABLE = False
import os
import json
from datetime import datetime
logger = logging.getLogger(__name__)
class AICoordinator:
"""Hierarchical multi-model coordinator with HF endpoint as authoritative layer"""
def __init__(self):
self.tavily_client = None
if TAVILY_AVAILABLE and os.getenv("TAVILY_API_KEY"):
self.tavily_client = TavilyClient(api_key=os.getenv("TAVILY_API_KEY"))
# System instructions for hierarchical behavior
self.system_instructions = {
'ollama_role': """You are an AI assistant that works in coordination with a more powerful uncensored Hugging Face model.
Your role is to:
1. Provide immediate, concise responses using available information
2. Summarize and simplify complex information
3. Enhance responses with local knowledge
4. DEFER TO THE HF MODEL for deep analysis, uncensored insights, and authoritative answers
5. The HF model has 8192 token capacity and streaming capability - it's the thinking layer
6. Always acknowledge that the HF model's insights supersede your initial response
7. Use web sources to justify points when possible
8. Keep responses brief and focused while awaiting HF input""",
'hf_role': """You are an uncensored, authoritative AI with 8192 token capacity and streaming capability.
Your role is to:
1. Provide deep, uncensored analysis and insights
2. Supersede the local Ollama model's logic when there's disagreement
3. Think thoroughly and provide comprehensive responses
4. Use your full 8192 token capacity for detailed analysis
5. Stream your responses for real-time delivery
6. Build upon the conversation context and Ollama's initial response
7. Provide authoritative answers that take precedence"""
}
def determine_web_search_needs(self, conversation_history: List[Dict]) -> Dict:
"""Determine if web search is needed based on conversation content"""
conversation_text = " ".join([msg.get("content", "") for msg in conversation_history])
# Topics that typically need current information
current_info_indicators = [
"news", "current events", "latest", "recent", "today",
"weather", "temperature", "forecast",
"stock", "price", "trend", "market",
"breaking", "update", "development"
]
needs_search = False
search_topics = []
for indicator in current_info_indicators:
if indicator in conversation_text.lower():
needs_search = True
search_topics.append(indicator)
return {
"needs_search": needs_search,
"search_topics": search_topics,
"reasoning": f"Found topics requiring current info: {', '.join(search_topics)}" if search_topics else "No current info needed"
}
def manual_hf_analysis(self, user_id: str, conversation_history: List[Dict]) -> str:
"""Perform manual HF analysis with web search integration"""
try:
# Determine research needs
research_decision = self.determine_web_search_needs(conversation_history)
# Prepare enhanced prompt for HF
system_prompt = f"""
You are a deep analysis expert joining an ongoing conversation.
Research Decision: {research_decision['reasoning']}
Please provide:
1. Deep insights on conversation themes
2. Research/web search needs (if any)
3. Strategic recommendations
4. Questions to explore further
Conversation History:
"""
# Add conversation history to messages
messages = [{"role": "system", "content": system_prompt}]
# Add recent conversation (last 15 messages for context)
for msg in conversation_history[-15:]:
messages.append({
"role": msg["role"],
"content": msg["content"]
})
# Get HF provider
from core.llm_factory import llm_factory
hf_provider = llm_factory.get_provider('huggingface')
if hf_provider:
# Generate deep analysis with full 8192 token capacity
response = hf_provider.generate("Deep analysis request", messages)
return response or "HF Expert analysis completed."
else:
return "❌ HF provider not available."
except Exception as e:
return f"❌ HF analysis failed: {str(e)}"
# Add this method to show HF engagement status
def get_hf_engagement_status(self) -> Dict:
"""Get current HF engagement status"""
return {
"hf_available": self._check_hf_availability(),
"web_search_configured": bool(self.tavily_client),
"research_needs_detected": False, # Will be determined per conversation
"last_hf_analysis": None # Track last analysis time
}
async def coordinate_hierarchical_conversation(self, user_id: str, user_query: str) -> AsyncGenerator[Dict, None]:
"""
Enhanced coordination with detailed tracking and feedback
"""
try:
# Get conversation history
session = session_manager.get_session(user_id)
conversation_history = session.get("conversation", []).copy()
yield {
'type': 'coordination_status',
'content': '🚀 Initiating hierarchical AI coordination...',
'details': {
'conversation_length': len(conversation_history),
'user_query_length': len(user_query)
}
}
# Step 1: Gather external data with detailed logging
yield {
'type': 'coordination_status',
'content': '🔍 Gathering external context...',
'details': {'phase': 'external_data_gathering'}
}
external_data = await self._gather_external_data(user_query)
# Log what external data was gathered
if external_data:
data_summary = []
if 'search_results' in external_data:
data_summary.append(f"Web search: {len(external_data['search_results'])} results")
if 'weather' in external_data:
data_summary.append("Weather data: available")
if 'current_datetime' in external_data:
data_summary.append(f"Time: {external_data['current_datetime']}")
yield {
'type': 'coordination_status',
'content': f'📊 External data gathered: {", ".join(data_summary)}',
'details': {'external_data_summary': data_summary}
}
# Step 2: Get initial Ollama response
yield {
'type': 'coordination_status',
'content': '🦙 Getting initial response from Ollama...',
'details': {'phase': 'ollama_response'}
}
ollama_response = await self._get_hierarchical_ollama_response(
user_query, conversation_history, external_data
)
# Send initial response with context info
yield {
'type': 'initial_response',
'content': ollama_response,
'details': {
'response_length': len(ollama_response),
'external_data_injected': bool(external_data)
}
}
# Step 3: Coordinate with HF endpoint
yield {
'type': 'coordination_status',
'content': '🤗 Engaging HF endpoint for deep analysis...',
'details': {'phase': 'hf_coordination'}
}
# Check HF availability
hf_available = self._check_hf_availability()
if hf_available:
# Show what context will be sent to HF
context_summary = {
'conversation_turns': len(conversation_history),
'ollama_response_length': len(ollama_response),
'external_data_items': len(external_data) if external_data else 0
}
yield {
'type': 'coordination_status',
'content': f'📋 HF context: {len(conversation_history)} conversation turns, Ollama response ({len(ollama_response)} chars)',
'details': context_summary
}
# Coordinate with HF
async for hf_chunk in self._coordinate_hierarchical_hf_response(
user_id, user_query, conversation_history,
external_data, ollama_response
):
yield hf_chunk
else:
yield {
'type': 'coordination_status',
'content': 'ℹ️ HF endpoint not available - using Ollama response',
'details': {'hf_available': False}
}
# Final coordination status
yield {
'type': 'coordination_status',
'content': '✅ Hierarchical coordination complete',
'details': {'status': 'complete'}
}
except Exception as e:
logger.error(f"Hierarchical coordination failed: {e}")
yield {
'type': 'coordination_status',
'content': f'❌ Coordination error: {str(e)}',
'details': {'error': str(e)}
}
async def _coordinate_hierarchical_hf_response(self, user_id: str, query: str,
history: List, external_data: Dict,
ollama_response: str) -> AsyncGenerator[Dict, None]:
"""Coordinate with HF endpoint as authoritative layer with streaming"""
try:
# Check and warm up HF endpoint if needed
hf_status = hf_monitor.check_endpoint_status()
if not hf_status['available']:
yield {'type': 'coordination_status', 'content': '⚡ Initializing HF endpoint (2-4 minutes)...'}
warmup_success = hf_monitor.handle_scale_to_zero()
if not warmup_success:
yield {'type': 'coordination_status', 'content': '❌ HF endpoint initialization failed'}
return
# Get HF provider
hf_provider = llm_factory.get_provider('huggingface')
if not hf_provider:
yield {'type': 'coordination_status', 'content': '❌ HF provider not available'}
return
# Prepare enhanced conversation for HF with hierarchical context
enhanced_history = history.copy()
# Add system instructions for HF
enhanced_history.insert(0, {
"role": "system",
"content": self.system_instructions['hf_role']
})
# Add context about the coordination
enhanced_history.append({
"role": "system",
"content": f"""Context for your authoritative analysis:
User Query: '{query}'
Ollama's Initial Response: '{ollama_response}'
External Data Available: {json.dumps(external_data, indent=2) if external_data else 'None'}
Please provide your deep, uncensored analysis that supersedes the local model's response.
Use your full 8192 token capacity for comprehensive thinking.
Stream your response for real-time delivery."""
})
# Add the user's latest query
enhanced_history.append({"role": "user", "content": query})
# Stream HF response with full 8192 token capacity
yield {'type': 'coordination_status', 'content': '🧠 HF endpoint thinking...'}
# Use streaming for real-time delivery
hf_response_stream = hf_provider.stream_generate(query, enhanced_history)
if hf_response_stream:
# Stream the response chunks
full_hf_response = ""
for chunk in hf_response_stream:
if chunk:
full_hf_response += chunk
yield {'type': 'hf_thinking', 'content': chunk}
# Final HF response
yield {'type': 'final_response', 'content': full_hf_response}
yield {'type': 'coordination_status', 'content': '🎯 HF analysis complete and authoritative'}
else:
yield {'type': 'coordination_status', 'content': '❌ HF response generation failed'}
except Exception as e:
logger.error(f"Hierarchical HF coordination failed: {e}")
yield {'type': 'coordination_status', 'content': f'❌ HF coordination error: {str(e)}'}
async def _get_hierarchical_ollama_response(self, query: str, history: List, external_data: Dict) -> str:
"""Get Ollama response with hierarchical awareness"""
try:
# Get Ollama provider
ollama_provider = llm_factory.get_provider('ollama')
if not ollama_provider:
raise Exception("Ollama provider not available")
# Prepare conversation with hierarchical context
enhanced_history = history.copy()
# Add system instruction for Ollama's role
enhanced_history.insert(0, {
"role": "system",
"content": self.system_instructions['ollama_role']
})
# Add external data context if available
if external_data:
context_parts = []
if 'search_answer' in external_data:
context_parts.append(f"Current information: {external_data['search_answer']}")
if 'weather' in external_data:
weather = external_data['weather']
context_parts.append(f"Current weather: {weather.get('temperature', 'N/A')}°C in {weather.get('city', 'Unknown')}")
if 'current_datetime' in external_data:
context_parts.append(f"Current time: {external_data['current_datetime']}")
if context_parts:
context_message = {
"role": "system",
"content": "Context: " + " | ".join(context_parts)
}
enhanced_history.insert(1, context_message) # Insert after role instruction
# Add the user's query
enhanced_history.append({"role": "user", "content": query})
# Generate response with awareness of HF's superior capabilities
response = ollama_provider.generate(query, enhanced_history)
# Add acknowledgment of HF's authority
if response:
return f"{response}\n\n*Note: A more comprehensive analysis from the uncensored HF model is being prepared...*"
else:
return "I'm processing your request... A deeper analysis is being prepared by the authoritative model."
except Exception as e:
logger.error(f"Hierarchical Ollama response failed: {e}")
return "I'm thinking about your question... Preparing a comprehensive response."
def _check_hf_availability(self) -> bool:
"""Check if HF endpoint is configured and available"""
try:
from utils.config import config
return bool(config.hf_token and config.hf_api_url)
except:
return False
async def _gather_external_data(self, query: str) -> Dict:
"""Gather external data from various sources"""
data = {}
# Tavily/DuckDuckGo search with justification focus
if self.tavily_client:
try:
search_result = self.tavily_client.search(
f"current information about {query}",
max_results=5, # More results for better justification
include_answer=True,
include_raw_content=True # For deeper analysis
)
data['search_results'] = search_result.get('results', [])
if search_result.get('answer'):
data['search_answer'] = search_result['answer']
# Store raw content for HF to analyze
data['raw_sources'] = [result.get('raw_content', '')[:1000] for result in search_result.get('results', [])[:3]]
except Exception as e:
logger.warning(f"Tavily search failed: {e}")
# Weather data
weather_keywords = ['weather', 'temperature', 'forecast', 'climate', 'rain', 'sunny']
if any(keyword in query.lower() for keyword in weather_keywords):
try:
location = self._extract_location(query) or "New York"
weather = weather_service.get_current_weather(location)
if weather:
data['weather'] = weather
except Exception as e:
logger.warning(f"Weather data failed: {e}")
# Current date/time
data['current_datetime'] = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
return data
def _extract_location(self, query: str) -> Optional[str]:
"""Extract location from query"""
locations = ['New York', 'London', 'Tokyo', 'Paris', 'Berlin', 'Sydney',
'Los Angeles', 'Chicago', 'Miami', 'Seattle', 'Boston',
'San Francisco', 'Toronto', 'Vancouver', 'Montreal']
for loc in locations:
if loc.lower() in query.lower():
return loc
return "New York" # Default
def get_coordination_status(self) -> Dict:
"""Get current coordination system status"""
return {
'tavily_available': self.tavily_client is not None,
'weather_available': weather_service.api_key is not None,
'web_search_enabled': self.tavily_client is not None,
'external_apis_configured': any([
weather_service.api_key,
os.getenv("TAVILY_API_KEY"),
os.getenv("NASA_API_KEY")
])
}
def get_recent_activities(self, user_id: str) -> Dict:
"""Get recent coordination activities for user"""
try:
session = session_manager.get_session(user_id)
coord_stats = session.get('ai_coordination', {})
return {
'last_request': coord_stats.get('last_coordination'),
'requests_processed': coord_stats.get('requests_processed', 0),
'ollama_responses': coord_stats.get('ollama_responses', 0),
'hf_responses': coord_stats.get('hf_responses', 0)
}
except:
return {}
# Global coordinator instance
coordinator = AICoordinator()