saap-plattform / backend /api /multi_agent_endpoints.py
Hwandji's picture
feat: initial HuggingFace Space deployment
4343907
#!/usr/bin/env python3
"""
Multi-Agent Communication API Endpoints for SAAP Platform
Provides REST API interface for multi-agent coordination and task delegation
"""
from fastapi import APIRouter, HTTPException, Depends
from typing import Dict, Any, Optional, List
import logging
from datetime import datetime
from pydantic import BaseModel
from services.multi_agent_coordinator import MultiAgentCoordinator, TaskPriority, get_coordinator
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Create router for multi-agent endpoints
multi_agent_router = APIRouter(prefix="/api/v1/multi-agent", tags=["Multi-Agent Communication"])
# 🔒 PRIVACY DETECTION HELPER
async def _determine_provider(user_message: str, provider_preference: Optional[str] = None) -> Dict[str, Any]:
"""
🔒 Autonomous Privacy-First Provider Selection
Analyzes user message for sensitive data and automatically selects appropriate provider:
- Colossus (internal): For sensitive/personal data (GDPR-protected)
- OpenRouter (external): For general queries (faster, cost-efficient)
Detection Categories:
- Medical: Patient data, diagnoses, symptoms, medications
- Financial: Account numbers, IBANs, credit cards, financial details
- Personal: Birthdates, addresses, phone numbers, ID numbers
- Legal: Legal cases, confidential information
Returns:
Dict with 'provider', 'reason', 'detected_categories', 'confidence'
"""
# Force provider if explicitly requested
if provider_preference and provider_preference != "auto":
logger.info(f"🔒 Provider forced by user: {provider_preference}")
return {
"provider": provider_preference,
"reason": "User preference",
"selected_provider": provider_preference,
"detected_categories": [],
"confidence": "explicit",
"auto_detected": False
}
# Sensitive keyword detection
message_lower = user_message.lower()
detected_categories = []
# Medical keywords (German + English)
medical_keywords = [
'patient', 'diagnose', 'diagnosis', 'krankheit', 'disease', 'symptom',
'arzt', 'doctor', 'medikament', 'medication', 'therapie', 'therapy',
'behandlung', 'treatment', 'gesundheit', 'health', 'krankenhaus', 'hospital',
'medizin', 'medicine', 'blut', 'blood', 'labor', 'test'
]
# Financial keywords
financial_keywords = [
'konto', 'account', 'iban', 'bic', 'kreditkarte', 'credit card',
'gehalt', 'salary', 'steuer', 'tax', 'bank', 'überweisung', 'transfer',
'rechnung', 'invoice', 'zahlung', 'payment', 'finanz', 'financial'
]
# Personal data keywords
personal_keywords = [
'geburtsdatum', 'birthdate', 'adresse', 'address', 'telefon', 'phone',
'ausweis', 'id', 'pass', 'passport', 'sozialversicherung', 'social security',
'privat', 'private', 'persönlich', 'personal'
]
# Legal keywords
legal_keywords = [
'vertrag', 'contract', 'klage', 'lawsuit', 'anwalt', 'lawyer',
'gericht', 'court', 'urteil', 'verdict', 'rechtlich', 'legal',
'compliance', 'datenschutz', 'gdpr', 'dsgvo'
]
# Check for sensitive keywords
if any(keyword in message_lower for keyword in medical_keywords):
detected_categories.append('medical')
if any(keyword in message_lower for keyword in financial_keywords):
detected_categories.append('financial')
if any(keyword in message_lower for keyword in personal_keywords):
detected_categories.append('personal')
if any(keyword in message_lower for keyword in legal_keywords):
detected_categories.append('legal')
# Pattern-based detection (PII)
import re
# Email pattern
if re.search(r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b', user_message):
detected_categories.append('email')
# Phone number pattern (German + International)
if re.search(r'\b(?:\+49|0)\s?\d{3,4}\s?\d{6,8}\b', user_message):
detected_categories.append('phone')
# IBAN pattern
if re.search(r'\b[A-Z]{2}\d{2}[A-Z0-9]{13,29}\b', user_message):
detected_categories.append('iban')
# Credit card pattern
if re.search(r'\b\d{4}[-\s]?\d{4}[-\s]?\d{4}[-\s]?\d{4}\b', user_message):
detected_categories.append('credit_card')
# Decision logic
if detected_categories:
# Sensitive data detected → Force Colossus
logger.warning(f"🔒 SENSITIVE DATA DETECTED: {detected_categories} → Colossus enforced")
return {
"provider": "colossus",
"reason": f"Sensitive data protection ({', '.join(detected_categories)})",
"selected_provider": "colossus",
"detected_categories": detected_categories,
"confidence": "high",
"auto_detected": True,
"privacy_level": "high"
}
else:
# No sensitive data → OpenRouter for speed/efficiency
logger.info("🌐 No sensitive data → OpenRouter selected")
return {
"provider": "openrouter",
"reason": "General query - optimized for speed",
"selected_provider": "openrouter",
"detected_categories": [],
"confidence": "high",
"auto_detected": True,
"privacy_level": "standard"
}
class MultiAgentChatRequest(BaseModel):
user_message: str
user_context: Optional[Dict[str, Any]] = None
preferred_agent: Optional[str] = None
task_priority: TaskPriority = TaskPriority.NORMAL
provider: Optional[str] = None # "auto", "colossus", "openrouter"
privacy_mode: Optional[str] = None # Alias for provider
class MultiAgentChatResponse(BaseModel):
success: bool
coordinator_response: str
delegated_agent: Optional[str] = None
specialist_response: Optional[str] = None
coordination_chain: List[str] = []
processing_time: float = 0.0
workflow_type: str = "single_agent"
task_id: Optional[str] = None
cost_info: Optional[Dict[str, Any]] = None
privacy_protection: Optional[Dict[str, Any]] = None # Privacy info
error: Optional[str] = None
@multi_agent_router.post("/chat", response_model=MultiAgentChatResponse)
async def multi_agent_chat(
request: MultiAgentChatRequest,
coordinator: MultiAgentCoordinator = Depends(get_coordinator)
):
"""
🤖 Multi-Agent Chat Endpoint - Jane Alesi Master Coordinator
Automatically analyzes user intent and either:
1. Handles request directly (Jane as coordinator)
2. Delegates to appropriate specialist agent
3. Orchestrates multi-agent workflow for complex tasks
Examples:
- "Entwickle eine Python App" → Jane delegates to John Alesi (Development)
- "Medizinische Beratung für Diabetes" → Jane delegates to Lara Alesi (Medical)
- "Legal Compliance Check" → Jane delegates to Justus Alesi (Legal)
- "SAAP Platform Status" → Jane handles directly as Coordinator
"""
start_time = datetime.now()
try:
logger.info(f"🤖 Multi-Agent Chat Request: {request.user_message[:100]}...")
# 🔒 PRIVACY-FIRST: Determine provider based on sensitivity
selected_provider = await _determine_provider(
user_message=request.user_message,
provider_preference=request.provider or request.privacy_mode
)
logger.info(f"🔒 Provider selection: {selected_provider['provider']} (reason: {selected_provider['reason']})")
# Execute multi-agent coordination with provider selection
coordination_result = await coordinator.coordinate_multi_agent_task(
user_message=request.user_message,
user_context=request.user_context or {},
provider=selected_provider['provider'] # Pass provider to coordinator
)
# Add privacy protection info to result
coordination_result['privacy_protection'] = selected_provider
processing_time = (datetime.now() - start_time).total_seconds()
if coordination_result.get("success", False):
# Successful coordination
workflow_type = coordination_result.get("workflow_type", "single_agent")
if workflow_type == "multi_agent":
# Complex multi-agent workflow
specialists = coordination_result.get("specialists", [])
workflow_steps = coordination_result.get("workflow_steps", [])
# Build coordination chain
coordination_chain = ["jane_alesi"] # Jane always starts
coordination_chain.extend(specialists)
# Get final response from synthesis step
coordinator_response = coordination_result.get("final_response", "Multi-agent workflow completed successfully.")
# Get specialist response (first specialist for simplicity)
specialist_response = None
delegated_agent = None
if workflow_steps:
for step in workflow_steps:
if step.get("step") == "specialist_analysis":
delegated_agent = step.get("agent")
specialist_response = step.get("result", {}).get("response", "Specialist analysis completed.")
break
logger.info(f"✅ Multi-Agent Workflow: {len(workflow_steps)} steps, {len(specialists)} specialists")
return MultiAgentChatResponse(
success=True,
coordinator_response=coordinator_response,
delegated_agent=delegated_agent,
specialist_response=specialist_response,
coordination_chain=coordination_chain,
processing_time=processing_time,
workflow_type="multi_agent",
task_id=coordination_result.get("task_id"),
cost_info={
"total_cost": 0.0, # Multi-agent coordination is free
"task_count": coordination_result.get("task_count", 1),
"agents_involved": len(coordination_chain)
},
privacy_protection=coordination_result.get('privacy_protection')
)
else:
# Single agent delegation
primary_agent = coordination_result.get("primary_agent", "jane_alesi")
response_text = coordination_result.get("response", "Task completed successfully.")
# Determine coordination chain
coordination_chain = ["jane_alesi"] # Jane analyzes intent
if primary_agent != "jane_alesi":
coordination_chain.append(primary_agent) # Delegate to specialist
coordination_chain.append("jane_alesi") # Jane provides final coordination
logger.info(f"✅ Single Agent Delegation: jane_alesi → {primary_agent}")
return MultiAgentChatResponse(
success=True,
coordinator_response=f"Als Master Coordinatorin habe ich deinen Request analysiert und {'direkt bearbeitet' if primary_agent == 'jane_alesi' else f'an {primary_agent} delegiert'}.",
delegated_agent=primary_agent if primary_agent != "jane_alesi" else None,
specialist_response=response_text if primary_agent != "jane_alesi" else None,
coordination_chain=coordination_chain,
processing_time=processing_time,
workflow_type="single_agent",
task_id=coordination_result.get("task_id"),
cost_info={
"total_cost": 0.0,
"agents_involved": len(coordination_chain)
},
privacy_protection=coordination_result.get('privacy_protection')
)
else:
# Coordination failed
error_msg = coordination_result.get("error", "Unknown coordination error")
logger.error(f"❌ Multi-Agent Coordination failed: {error_msg}")
return MultiAgentChatResponse(
success=False,
coordinator_response="Als Master Coordinatorin konnte ich deinen Request leider nicht erfolgreich bearbeiten.",
processing_time=processing_time,
error=error_msg
)
except Exception as e:
processing_time = (datetime.now() - start_time).total_seconds()
logger.error(f"❌ Multi-Agent Chat API Error: {e}")
return MultiAgentChatResponse(
success=False,
coordinator_response="Entschuldigung, es ist ein technischer Fehler im Multi-Agent System aufgetreten.",
processing_time=processing_time,
error=str(e)
)
@multi_agent_router.get("/status")
async def get_multi_agent_status(
coordinator: MultiAgentCoordinator = Depends(get_coordinator)
):
"""
Get current multi-agent coordination status and statistics
"""
try:
stats = await coordinator.get_coordination_stats()
return {
"status": "active",
"coordinator": "jane_alesi",
"available_specialists": [
{"id": "john_alesi", "name": "John Alesi", "specialization": "Development"},
{"id": "lara_alesi", "name": "Lara Alesi", "specialization": "Medical"},
{"id": "justus_alesi", "name": "Justus Alesi", "specialization": "Legal"},
{"id": "theo_alesi", "name": "Theo Alesi", "specialization": "Finance"},
{"id": "leon_alesi", "name": "Leon Alesi", "specialization": "System"},
{"id": "luna_alesi", "name": "Luna Alesi", "specialization": "Coaching"}
],
"coordination_stats": stats,
"features": {
"intent_analysis": True,
"automatic_delegation": True,
"multi_agent_workflows": True,
"real_time_coordination": True,
"task_orchestration": True
},
"timestamp": datetime.now().isoformat()
}
except Exception as e:
logger.error(f"❌ Multi-Agent Status Error: {e}")
raise HTTPException(status_code=500, detail=f"Status check failed: {str(e)}")
@multi_agent_router.get("/capabilities")
async def get_agent_capabilities(
coordinator: MultiAgentCoordinator = Depends(get_coordinator)
):
"""
Get detailed agent capabilities for intelligent task delegation
"""
try:
capabilities = {}
for agent_id, agent_caps in coordinator.agent_capabilities.items():
capabilities[agent_id] = {
"agent_name": {
"jane_alesi": "Jane Alesi - Master Coordinator",
"john_alesi": "John Alesi - Software Developer",
"lara_alesi": "Lara Alesi - Medical Expert",
"justus_alesi": "Justus Alesi - Legal Expert",
"theo_alesi": "Theo Alesi - Financial Analyst",
"leon_alesi": "Leon Alesi - System Administrator",
"luna_alesi": "Luna Alesi - Coaching Specialist"
}.get(agent_id, agent_id),
"capabilities": [
{
"name": cap.name,
"description": cap.description,
"keywords": cap.keywords,
"complexity_level": cap.complexity_level
}
for cap in agent_caps
],
"specialization": {
"jane_alesi": "Coordination & Architecture",
"john_alesi": "Software Development",
"lara_alesi": "Medical Analysis",
"justus_alesi": "Legal Compliance",
"theo_alesi": "Financial Analysis",
"leon_alesi": "System Administration",
"luna_alesi": "Coaching & Process"
}.get(agent_id, "General")
}
return {
"total_agents": len(capabilities),
"coordinator": "jane_alesi",
"specialists_count": len(capabilities) - 1,
"capabilities": capabilities,
"timestamp": datetime.now().isoformat()
}
except Exception as e:
logger.error(f"❌ Agent Capabilities Error: {e}")
raise HTTPException(status_code=500, detail=f"Capabilities retrieval failed: {str(e)}")
@multi_agent_router.get("/workload/{agent_id}")
async def get_agent_workload(
agent_id: str,
coordinator: MultiAgentCoordinator = Depends(get_coordinator)
):
"""
Get current workload and task statistics for a specific agent
"""
try:
workload = await coordinator.get_agent_workload(agent_id)
return workload
except Exception as e:
logger.error(f"❌ Agent Workload Error for {agent_id}: {e}")
raise HTTPException(status_code=500, detail=f"Workload check failed: {str(e)}")