PraisonAI 3.0.0__py3-none-any.whl

praisonai/cli/features/router.py

@@ -0,0 +1,466 @@
+"""
+Router Handler for CLI.
+
+Provides smart model selection based on task complexity.
+Usage: praisonai "Complex task" --router
+
+Supports custom model configuration from agents.yaml/workflow.yaml:
+
+models:
+  gpt-4o-mini:
+    provider: openai
+    complexity: [simple, moderate]
+    cost_per_1k: 0.00075
+    capabilities: [text, function-calling]
+"""
+
+from typing import Any, Dict, List, Optional, Tuple
+from dataclasses import dataclass
+from enum import IntEnum
+from .base import FlagHandler
+
+
+class TaskComplexity(IntEnum):
+    """Enum for task complexity levels."""
+    SIMPLE = 1
+    MODERATE = 2
+    COMPLEX = 3
+    VERY_COMPLEX = 4
+
+
+@dataclass
+class ModelProfile:
+    """Profile for an LLM model with its characteristics."""
+    name: str
+    provider: str
+    complexity_range: Tuple[TaskComplexity, TaskComplexity]
+    cost_per_1k_tokens: float
+    capabilities: List[str]
+    context_window: int = 128000
+    supports_tools: bool = True
+    supports_streaming: bool = True
+    strengths: Optional[List[str]] = None
+
+
+class RouterHandler(FlagHandler):
+    """
+    Handler for --router flag.
+    
+    Automatically selects the best model based on task complexity.
+    
+    Example:
+        praisonai "Simple question" --router
+        praisonai "Complex analysis" --router
+    """
+    
+    # Default model tiers for routing (used when no custom models configured)
+    # Note: Use full model names for API compatibility
+    DEFAULT_MODEL_TIERS = {
+        'simple': ['gpt-4o-mini', 'anthropic/claude-3-haiku-20240307', 'gemini/gemini-1.5-flash'],
+        'medium': ['gpt-4o', 'anthropic/claude-3-5-sonnet-20241022', 'gemini/gemini-1.5-pro'],
+        'complex': ['gpt-4-turbo', 'anthropic/claude-3-opus-20240229', 'o1-preview'],
+    }
+    
+    # Default model profiles
+    DEFAULT_MODELS: List[ModelProfile] = [
+        ModelProfile(
+            name="gpt-4o-mini",
+            provider="openai",
+            complexity_range=(TaskComplexity.SIMPLE, TaskComplexity.MODERATE),
+            cost_per_1k_tokens=0.00075,
+            capabilities=["text", "function-calling"],
+            context_window=128000
+        ),
+        ModelProfile(
+            name="gpt-4o",
+            provider="openai",
+            complexity_range=(TaskComplexity.MODERATE, TaskComplexity.COMPLEX),
+            cost_per_1k_tokens=0.0075,
+            capabilities=["text", "vision", "function-calling"],
+            context_window=128000
+        ),
+        ModelProfile(
+            name="anthropic/claude-3-haiku-20240307",
+            provider="anthropic",
+            complexity_range=(TaskComplexity.SIMPLE, TaskComplexity.MODERATE),
+            cost_per_1k_tokens=0.0008,
+            capabilities=["text", "function-calling"],
+            context_window=200000
+        ),
+        ModelProfile(
+            name="anthropic/claude-3-5-sonnet-20241022",
+            provider="anthropic",
+            complexity_range=(TaskComplexity.MODERATE, TaskComplexity.VERY_COMPLEX),
+            cost_per_1k_tokens=0.009,
+            capabilities=["text", "vision", "function-calling"],
+            context_window=200000,
+            strengths=["reasoning", "code-generation", "analysis"]
+        ),
+        ModelProfile(
+            name="gemini/gemini-1.5-flash",
+            provider="google",
+            complexity_range=(TaskComplexity.SIMPLE, TaskComplexity.MODERATE),
+            cost_per_1k_tokens=0.000125,
+            capabilities=["text", "vision", "function-calling"],
+            context_window=1048576
+        ),
+        ModelProfile(
+            name="gemini/gemini-1.5-pro",
+            provider="google",
+            complexity_range=(TaskComplexity.COMPLEX, TaskComplexity.VERY_COMPLEX),
+            cost_per_1k_tokens=0.00625,
+            capabilities=["text", "vision", "function-calling"],
+            context_window=2097152
+        ),
+    ]
+    
+    def __init__(self, verbose: bool = False):
+        """Initialize RouterHandler with model routing support."""
+        super().__init__(verbose=verbose)
+        self._model_router: Optional[List[ModelProfile]] = None
+        self._custom_models: Dict[str, ModelProfile] = {}
+        self._cost_threshold: Optional[float] = None
+        self._model_tiers = self.DEFAULT_MODEL_TIERS.copy()
+    
+    @property
+    def model_router(self) -> List[ModelProfile]:
+        """Get the list of available model profiles."""
+        if self._model_router is None:
+            return self.DEFAULT_MODELS.copy()
+        return self._model_router
+    
+    @property
+    def feature_name(self) -> str:
+        return "router"
+    
+    @property
+    def flag_name(self) -> str:
+        return "router"
+    
+    @property
+    def flag_help(self) -> str:
+        return "Auto-select best model based on task complexity"
+    
+    def check_dependencies(self) -> Tuple[bool, str]:
+        """Router is built-in, always available."""
+        return True, ""
+    
+    def analyze_complexity(self, prompt: str) -> str:
+        """
+        Analyze prompt complexity.
+        
+        Args:
+            prompt: The task prompt
+            
+        Returns:
+            Complexity level: 'simple', 'medium', or 'complex'
+        """
+        # Simple heuristics for complexity
+        prompt_lower = prompt.lower()
+        word_count = len(prompt.split())
+        
+        # Complex indicators
+        complex_keywords = [
+            'analyze', 'research', 'comprehensive', 'detailed',
+            'compare', 'evaluate', 'synthesize', 'multi-step',
+            'code review', 'architecture', 'design pattern',
+            'optimize', 'debug', 'refactor'
+        ]
+        
+        # Simple indicators
+        simple_keywords = [
+            'what is', 'define', 'list', 'name', 'when',
+            'where', 'who', 'simple', 'quick', 'brief'
+        ]
+        
+        complex_count = sum(1 for kw in complex_keywords if kw in prompt_lower)
+        simple_count = sum(1 for kw in simple_keywords if kw in prompt_lower)
+        
+        # Determine complexity
+        if complex_count >= 2 or word_count > 100:
+            return 'complex'
+        elif simple_count >= 2 or word_count < 20:
+            return 'simple'
+        else:
+            return 'medium'
+    
+    def select_model(self, prompt: str, preferred_provider: str = None) -> str:
+        """
+        Select the best model for the prompt.
+        
+        Args:
+            prompt: The task prompt
+            preferred_provider: Optional preferred provider (openai, anthropic, google)
+            
+        Returns:
+            Selected model name
+        """
+        complexity = self.analyze_complexity(prompt)
+        
+        # If custom models are loaded, use ModelProfile-based selection
+        if self._custom_models:
+            return self._select_from_profiles(prompt, complexity, preferred_provider)
+        
+        # Otherwise use tier-based selection
+        models = self._model_tiers.get(complexity, self._model_tiers['medium'])
+        
+        # Filter by provider if specified
+        if preferred_provider:
+            provider_prefixes = {
+                'openai': ['gpt-', 'o1-'],
+                'anthropic': ['anthropic/', 'claude-'],
+                'google': ['gemini/', 'gemini-']
+            }
+            prefixes = provider_prefixes.get(preferred_provider.lower(), [])
+            filtered = [m for m in models if any(p in m for p in prefixes)]
+            if filtered:
+                models = filtered
+        
+        # Apply cost threshold if set
+        if self._cost_threshold:
+            filtered = [m for m in models if self.get_model_cost(m) <= self._cost_threshold]
+            if filtered:
+                models = filtered
+        
+        selected = models[0] if models else 'gpt-4o-mini'
+        
+        self.print_status(f"🎯 Task complexity: {complexity}", "info")
+        self.print_status(f"🤖 Selected model: {selected}", "success")
+        
+        return selected
+    
+    def _select_from_profiles(
+        self, 
+        prompt: str, 
+        complexity: str, 
+        preferred_provider: str = None
+    ) -> str:
+        """Select model from custom ModelProfile objects."""
+        # Map string complexity to TaskComplexity enum
+        complexity_map = {
+            'simple': TaskComplexity.SIMPLE,
+            'medium': TaskComplexity.MODERATE,
+            'complex': TaskComplexity.COMPLEX,
+        }
+        task_complexity = complexity_map.get(complexity, TaskComplexity.MODERATE)
+        
+        # Get all models that support this complexity level
+        candidates = []
+        for name, profile in self._custom_models.items():
+            min_c, max_c = profile.complexity_range
+            if min_c.value <= task_complexity.value <= max_c.value:
+                candidates.append(profile)
+        
+        if not candidates:
+            return 'gpt-4o-mini'
+        
+        # Filter by provider if specified
+        if preferred_provider:
+            filtered = [m for m in candidates if m.provider == preferred_provider]
+            if filtered:
+                candidates = filtered
+        
+        # Filter by cost threshold if set
+        if self._cost_threshold:
+            filtered = [m for m in candidates if m.cost_per_1k_tokens <= self._cost_threshold]
+            if filtered:
+                candidates = filtered
+        
+        # Sort by cost (cheapest first)
+        candidates.sort(key=lambda m: m.cost_per_1k_tokens)
+        
+        return candidates[0].name if candidates else 'gpt-4o-mini'
+    
+    def apply_to_agent_config(self, config: Dict[str, Any], flag_value: Any) -> Dict[str, Any]:
+        """
+        Apply router configuration.
+        
+        Args:
+            config: Agent configuration dictionary
+            flag_value: Boolean or dict with provider preference
+            
+        Returns:
+            Modified configuration with selected model
+        """
+        if not flag_value:
+            return config
+        
+        # Get prompt from config if available
+        prompt = config.get('prompt', config.get('goal', ''))
+        
+        preferred_provider = None
+        if isinstance(flag_value, dict):
+            preferred_provider = flag_value.get('provider')
+        
+        if prompt:
+            selected_model = self.select_model(prompt, preferred_provider)
+            config['llm'] = selected_model
+        
+        config['use_router'] = True
+        return config
+    
+    def execute(self, prompt: str = None, provider: str = None, **kwargs) -> str:
+        """
+        Execute model selection.
+        
+        Args:
+            prompt: Task prompt
+            provider: Preferred provider
+            
+        Returns:
+            Selected model name
+        """
+        if not prompt:
+            return 'gpt-4o-mini'  # Default
+        
+        return self.select_model(prompt, provider)
+    
+    # =========================================================================
+    # Custom Model Configuration Methods
+    # =========================================================================
+    
+    def yaml_to_model_profile(self, name: str, config: Dict[str, Any]) -> ModelProfile:
+        """
+        Convert YAML model configuration to ModelProfile object.
+        
+        Args:
+            name: Model name
+            config: Model configuration from YAML
+            
+        Returns:
+            ModelProfile object
+        """
+        # Parse complexity strings to TaskComplexity enum
+        complexity_list = config.get('complexity', ['moderate'])
+        complexity_map = {
+            'simple': TaskComplexity.SIMPLE,
+            'moderate': TaskComplexity.MODERATE,
+            'complex': TaskComplexity.COMPLEX,
+            'very_complex': TaskComplexity.VERY_COMPLEX,
+        }
+        
+        complexities = [complexity_map.get(c.lower(), TaskComplexity.MODERATE) for c in complexity_list]
+        min_complexity = min(complexities, key=lambda x: x.value)
+        max_complexity = max(complexities, key=lambda x: x.value)
+        
+        return ModelProfile(
+            name=name,
+            provider=config.get('provider', 'unknown'),
+            complexity_range=(min_complexity, max_complexity),
+            cost_per_1k_tokens=config.get('cost_per_1k', 0.001),
+            capabilities=config.get('capabilities', ['text']),
+            context_window=config.get('context_window', 128000),
+            supports_tools=config.get('supports_tools', True),
+            supports_streaming=config.get('supports_streaming', True),
+            strengths=config.get('strengths', None)
+        )
+    
+    def load_models_from_config(
+        self, 
+        models_config: Dict[str, Dict[str, Any]], 
+        merge_with_defaults: bool = False
+    ) -> None:
+        """
+        Load custom models from YAML configuration.
+        
+        Args:
+            models_config: Dictionary of model configurations from YAML
+            merge_with_defaults: If True, merge with default models
+        """
+        # Start with defaults if merging
+        if merge_with_defaults:
+            for profile in self.DEFAULT_MODELS:
+                self._custom_models[profile.name] = profile
+        
+        # Add custom models
+        for name, config in models_config.items():
+            profile = self.yaml_to_model_profile(name, config)
+            self._custom_models[name] = profile
+        
+        # Rebuild model tiers from profiles
+        self._rebuild_model_tiers()
+    
+    def _rebuild_model_tiers(self) -> None:
+        """Rebuild MODEL_TIERS from custom model profiles."""
+        self._model_tiers = {
+            'simple': [],
+            'medium': [],
+            'complex': [],
+        }
+        
+        for name, profile in self._custom_models.items():
+            min_c, max_c = profile.complexity_range
+            
+            if min_c.value <= TaskComplexity.SIMPLE.value <= max_c.value:
+                self._model_tiers['simple'].append(name)
+            if min_c.value <= TaskComplexity.MODERATE.value <= max_c.value:
+                self._model_tiers['medium'].append(name)
+            if min_c.value <= TaskComplexity.COMPLEX.value <= max_c.value:
+                self._model_tiers['complex'].append(name)
+        
+        # Sort by cost (cheapest first)
+        for tier in self._model_tiers:
+            self._model_tiers[tier].sort(
+                key=lambda n: self._custom_models[n].cost_per_1k_tokens
+            )
+    
+    def get_available_models(self) -> List[str]:
+        """
+        Get list of available model names.
+        
+        Returns:
+            List of model names
+        """
+        if self._custom_models:
+            return list(self._custom_models.keys())
+        return [p.name for p in self.DEFAULT_MODELS]
+    
+    def get_model_cost(self, model_name: str) -> float:
+        """
+        Get cost per 1k tokens for a model.
+        
+        Args:
+            model_name: Name of the model
+            
+        Returns:
+            Cost per 1k tokens
+        """
+        # Check custom models first
+        if model_name in self._custom_models:
+            return self._custom_models[model_name].cost_per_1k_tokens
+        
+        # Check default models
+        for profile in self.DEFAULT_MODELS:
+            if profile.name == model_name:
+                return profile.cost_per_1k_tokens
+        
+        # Default cost for unknown models
+        return 0.01
+    
+    def set_cost_threshold(self, threshold: float) -> None:
+        """
+        Set maximum cost per 1k tokens for model selection.
+        
+        Args:
+            threshold: Maximum cost per 1k tokens
+        """
+        self._cost_threshold = threshold
+    
+    def get_model_profile(self, model_name: str) -> Optional[ModelProfile]:
+        """
+        Get ModelProfile for a model.
+        
+        Args:
+            model_name: Name of the model
+            
+        Returns:
+            ModelProfile or None if not found
+        """
+        if model_name in self._custom_models:
+            return self._custom_models[model_name]
+        
+        for profile in self.DEFAULT_MODELS:
+            if profile.name == model_name:
+                return profile
+        
+        return None