saap-plattform / backend /api /openrouter_client.py
Hwandji's picture
feat: initial HuggingFace Space deployment
4343907
"""
OpenRouter API Client for SAAP
Provides OpenAI-compatible interface with cost tracking and performance metrics
"""
import asyncio
import logging
import time
import os
from typing import Dict, List, Optional, Any, Tuple
from datetime import datetime
import aiohttp
import json
from dataclasses import dataclass
from dotenv import load_dotenv
# Load environment variables
load_dotenv()
logger = logging.getLogger(__name__)
@dataclass
class OpenRouterResponse:
"""OpenRouter API response with cost tracking"""
success: bool
content: Optional[str] = None
error: Optional[str] = None
response_time: float = 0.0
tokens_used: int = 0
input_tokens: int = 0
output_tokens: int = 0
cost_usd: float = 0.0
model: str = ""
provider: str = "openrouter"
timestamp: datetime = None
def __post_init__(self):
if self.timestamp is None:
self.timestamp = datetime.utcnow()
def to_dict(self) -> Dict[str, Any]:
"""Convert to dictionary for logging and API responses"""
return {
"success": self.success,
"content": self.content,
"error": self.error,
"response_time": self.response_time,
"tokens_used": self.tokens_used,
"input_tokens": self.input_tokens,
"output_tokens": self.output_tokens,
"cost_usd": self.cost_usd,
"model": self.model,
"provider": self.provider,
"timestamp": self.timestamp.isoformat(),
"cost_efficiency": f"${self.cost_usd:.6f} ({self.tokens_used} tokens, {self.response_time:.1f}s)" if self.success else "N/A"
}
class OpenRouterClient:
"""
OpenRouter API Client with Cost Optimization for SAAP
Features:
- OpenAI-compatible API interface
- Agent-specific model selection (Jane: GPT-4o-mini, John: Claude-3.5-Sonnet, etc.)
- Cost tracking and budget management
- Performance monitoring and fallback models
- Async/await support for high-performance integration
"""
def __init__(self, api_key: str, base_url: str = "https://openrouter.ai/api/v1"):
self.api_key = api_key
self.base_url = base_url.rstrip('/')
self.session: Optional[aiohttp.ClientSession] = None
self.daily_budget = 10.0 # $10/day default
self.current_daily_cost = 0.0
self.cost_alert_threshold = 0.8 # 80% of budget
# πŸš€ PERFORMANCE OPTIMIZATION: Reduced token limits for faster responses
# Phase 1.3 Quick Win: 40-50% token reduction = 0.5-1s faster per request
# Agent-specific model configurations with cost data
self.agent_models = {
"jane_alesi": {
"model": "openai/gpt-4o-mini",
"max_tokens": 400, # Reduced from 800 (-50%)
"temperature": 0.7,
"cost_per_1m_input": 0.15, # $0.15/1M tokens
"cost_per_1m_output": 0.60, # $0.60/1M tokens
"description": "Efficient coordination and management"
},
"john_alesi": {
"model": "anthropic/claude-3-5-sonnet-20241022",
"max_tokens": 600, # Reduced from 1200 (-50%)
"temperature": 0.5,
"cost_per_1m_input": 3.00, # $3.00/1M tokens
"cost_per_1m_output": 15.00, # $15.00/1M tokens
"description": "Advanced code generation and development"
},
"lara_alesi": {
"model": "openai/gpt-4o-mini",
"max_tokens": 500, # Reduced from 1000 (-50%)
"temperature": 0.3,
"cost_per_1m_input": 0.15, # $0.15/1M tokens
"cost_per_1m_output": 0.60, # $0.60/1M tokens
"description": "Precise medical and analytical tasks"
},
"fallback": {
"model": "meta-llama/llama-3.2-3b-instruct:free",
"max_tokens": 600,
"temperature": 0.7,
"cost_per_1m_input": 0.0, # FREE
"cost_per_1m_output": 0.0, # FREE
"description": "Free backup model for budget protection"
}
}
logger.info(f"🌐 OpenRouter Client initialized with {len(self.agent_models)} agent models")
logger.info(f"πŸ’° Daily budget: ${self.daily_budget}, Alert threshold: {self.cost_alert_threshold*100}%")
async def __aenter__(self):
"""Async context manager entry"""
self.session = aiohttp.ClientSession(
timeout=aiohttp.ClientTimeout(total=60),
headers={
"Authorization": f"Bearer {self.api_key}",
"Content-Type": "application/json",
"HTTP-Referer": "https://saap.satware.ai", # Optional: your app URL
"X-Title": "SAAP Agent Platform" # Optional: app title
}
)
logger.info("πŸ”Œ OpenRouter session created")
return self
async def __aexit__(self, exc_type, exc_val, exc_tb):
"""Async context manager exit"""
if self.session:
await self.session.close()
logger.info("πŸ”Œ OpenRouter session closed")
def get_model_config(self, agent_id: str) -> Dict[str, Any]:
"""Get model configuration for specific agent"""
return self.agent_models.get(agent_id, self.agent_models["fallback"])
def estimate_cost(self, message: str, agent_id: str) -> float:
"""Estimate request cost before sending"""
config = self.get_model_config(agent_id)
# Rough token estimation: ~4 characters per token
estimated_input_tokens = len(message) / 4
estimated_output_tokens = config["max_tokens"] * 0.5 # Assume 50% of max tokens
cost_usd = (
(estimated_input_tokens / 1_000_000) * config["cost_per_1m_input"] +
(estimated_output_tokens / 1_000_000) * config["cost_per_1m_output"]
)
return cost_usd
async def chat_completion(
self,
messages: List[Dict[str, str]],
agent_id: str,
max_tokens: Optional[int] = None,
temperature: Optional[float] = None
) -> OpenRouterResponse:
"""
Send chat completion request to OpenRouter with cost tracking
Args:
messages: List of message dicts with 'role' and 'content'
agent_id: Agent identifier for model selection
max_tokens: Override max tokens
temperature: Override temperature
Returns:
OpenRouterResponse with content, cost, and performance data
"""
if not self.session:
return OpenRouterResponse(
success=False,
error="OpenRouter client session not initialized - call async context manager",
model="",
provider="openrouter"
)
# Get agent-specific model config
config = self.get_model_config(agent_id)
model = config["model"]
# Budget check before expensive request
estimated_cost = self.estimate_cost(str(messages), agent_id)
if self.current_daily_cost + estimated_cost > self.daily_budget:
logger.warning(f"πŸ’Έ Daily budget would be exceeded - switching to free fallback")
config = self.agent_models["fallback"]
model = config["model"]
start_time = time.time()
# Prepare request payload
payload = {
"model": model,
"messages": messages,
"max_tokens": max_tokens or config["max_tokens"],
"temperature": temperature or config["temperature"],
"stream": False # We want complete responses for cost calculation
}
logger.info(f"πŸ“€ OpenRouter request: {agent_id} β†’ {model}")
try:
async with self.session.post(
f"{self.base_url}/chat/completions",
json=payload
) as response:
response_time = time.time() - start_time
if response.status == 200:
data = await response.json()
# Extract response content
content = ""
if "choices" in data and len(data["choices"]) > 0:
choice = data["choices"][0]
if "message" in choice and "content" in choice["message"]:
content = choice["message"]["content"]
# Extract token usage and calculate cost
usage = data.get("usage", {})
input_tokens = usage.get("prompt_tokens", 0)
output_tokens = usage.get("completion_tokens", 0)
total_tokens = usage.get("total_tokens", 0)
# Calculate actual cost
cost_usd = (
(input_tokens / 1_000_000) * config["cost_per_1m_input"] +
(output_tokens / 1_000_000) * config["cost_per_1m_output"]
)
# Update daily cost tracking
self.current_daily_cost += cost_usd
# Log cost alert if needed
if self.current_daily_cost >= (self.daily_budget * self.cost_alert_threshold):
logger.warning(f"⚠️ OpenRouter cost alert: ${self.current_daily_cost:.4f} / ${self.daily_budget} ({self.current_daily_cost/self.daily_budget*100:.1f}%)")
logger.info(f"βœ… OpenRouter success: {response_time:.2f}s, ${cost_usd:.6f}, {total_tokens} tokens")
return OpenRouterResponse(
success=True,
content=content,
response_time=response_time,
tokens_used=total_tokens,
input_tokens=input_tokens,
output_tokens=output_tokens,
cost_usd=cost_usd,
model=model,
provider="openrouter"
)
else:
error_text = await response.text()
error_msg = f"HTTP {response.status}: {error_text}"
# Handle rate limiting and payment errors with fallback
if response.status in [429, 402]:
logger.warning(f"⚠️ OpenRouter limit reached: {error_msg}")
if config != self.agent_models["fallback"]:
logger.info("πŸ”„ Attempting fallback to free model...")
return await self.chat_completion(messages, "fallback", max_tokens, temperature)
logger.error(f"❌ OpenRouter API error: {error_msg}")
return OpenRouterResponse(
success=False,
error=error_msg,
response_time=response_time,
model=model,
provider="openrouter"
)
except asyncio.TimeoutError:
error_msg = f"Request timeout after {time.time() - start_time:.1f}s"
logger.error(f"❌ OpenRouter timeout: {error_msg}")
return OpenRouterResponse(
success=False,
error=error_msg,
response_time=time.time() - start_time,
model=model,
provider="openrouter"
)
except Exception as e:
error_msg = f"OpenRouter request failed: {str(e)}"
logger.error(f"❌ OpenRouter error: {error_msg}")
return OpenRouterResponse(
success=False,
error=error_msg,
response_time=time.time() - start_time,
model=model,
provider="openrouter"
)
async def health_check(self) -> Dict[str, Any]:
"""Check OpenRouter API health and available models"""
if not self.session:
return {
"status": "unhealthy",
"error": "Session not initialized"
}
try:
# Test with a simple completion
test_messages = [{"role": "user", "content": "Reply with just 'OK' to confirm API connection."}]
result = await self.chat_completion(test_messages, "fallback") # Use free model for health check
return {
"status": "healthy" if result.success else "unhealthy",
"response_time": result.response_time,
"error": result.error if not result.success else None,
"daily_cost": self.current_daily_cost,
"budget_remaining": max(0, self.daily_budget - self.current_daily_cost),
"available_models": len(self.agent_models),
"timestamp": datetime.utcnow().isoformat()
}
except Exception as e:
return {
"status": "error",
"error": str(e),
"timestamp": datetime.utcnow().isoformat()
}
def get_cost_summary(self) -> Dict[str, Any]:
"""Get current cost and budget status"""
return {
"daily_cost_usd": round(self.current_daily_cost, 4),
"daily_budget_usd": self.daily_budget,
"budget_used_percent": round(self.current_daily_cost / self.daily_budget * 100, 1),
"budget_remaining_usd": max(0, self.daily_budget - self.current_daily_cost),
"alert_threshold_percent": self.cost_alert_threshold * 100,
"cost_alert_active": self.current_daily_cost >= (self.daily_budget * self.cost_alert_threshold),
"agent_models_available": list(self.agent_models.keys()),
"timestamp": datetime.utcnow().isoformat()
}
def reset_daily_costs(self):
"""Reset daily cost tracking (call at midnight)"""
yesterday_cost = self.current_daily_cost
self.current_daily_cost = 0.0
logger.info(f"πŸ“… OpenRouter daily cost reset - Yesterday: ${yesterday_cost:.4f}")
if __name__ == "__main__":
# Demo OpenRouter integration
async def demo_openrouter():
api_key = os.getenv("OPENROUTER_API_KEY", "")
if not api_key:
print("❌ Error: OPENROUTER_API_KEY environment variable not set")
print("Please add OPENROUTER_API_KEY to backend/.env file")
return
async with OpenRouterClient(api_key) as client:
print("🌐 OpenRouter Client Demo")
# Health check
health = await client.health_check()
print(f"Health: {health['status']}")
if health["status"] == "healthy":
# Test different agents
for agent in ["jane_alesi", "john_alesi", "fallback"]:
config = client.get_model_config(agent)
print(f"\nπŸ€– Testing {agent} - Model: {config['model']}")
messages = [
{"role": "user", "content": f"Hello! I'm testing the {agent} agent. Please respond briefly."}
]
result = await client.chat_completion(messages, agent)
if result.success:
print(f"βœ… Response: {result.content[:100]}...")
print(f"πŸ’° Cost: ${result.cost_usd:.6f}")
print(f"⏱️ Time: {result.response_time:.2f}s")
print(f"πŸ”’ Tokens: {result.tokens_used}")
else:
print(f"❌ Error: {result.error}")
# Cost summary
print(f"\nπŸ’° Cost Summary: {client.get_cost_summary()}")
asyncio.run(demo_openrouter())