signalwire-agents 0.1.6__py3-none-any.whl → 1.0.7__py3-none-any.whl

signalwire_agents/agents/bedrock.py

@@ -0,0 +1,296 @@
+"""
+Copyright (c) 2025 SignalWire
+
+This file is part of the SignalWire AI Agents SDK.
+
+Licensed under the MIT License.
+See LICENSE file in the project root for full license information.
+"""
+
+"""
+Bedrock Agent - Amazon Bedrock voice-to-voice integration
+
+This module provides BedrockAgent, which extends AgentBase to support
+Amazon Bedrock's voice-to-voice model while maintaining compatibility
+with all SignalWire agent features like skills, POM, and SWAIG functions.
+"""
+
+import json
+from typing import Dict, List, Any, Optional, Union
+from signalwire_agents.core.agent_base import AgentBase
+from signalwire_agents.core.logging_config import get_logger
+
+logger = get_logger("bedrock_agent")
+
+
+class BedrockAgent(AgentBase):
+    """
+    Agent implementation for Amazon Bedrock voice-to-voice model
+    
+    This agent extends AgentBase to provide full compatibility with
+    SignalWire's agent ecosystem while using Amazon Bedrock as the
+    AI backend. It supports all standard agent features including:
+    - Prompt building with text and POM
+    - Skills and SWAIG functions
+    - Post-prompt functionality
+    - Dynamic configuration
+    
+    The main difference from the standard agent is that it generates
+    SWML with the "amazon_bedrock" verb instead of "ai".
+    """
+    
+    def __init__(
+        self,
+        name: str = "bedrock_agent",
+        route: str = "/bedrock",
+        system_prompt: Optional[str] = None,
+        voice_id: str = "matthew",
+        temperature: float = 0.7,
+        top_p: float = 0.9,
+        max_tokens: int = 1024,
+        **kwargs
+    ):
+        """
+        Initialize BedrockAgent
+        
+        Args:
+            name: Agent name
+            route: HTTP route for the agent
+            system_prompt: Initial system prompt (can be overridden with set_prompt)
+            voice_id: Bedrock voice ID (default: matthew)
+            temperature: Generation temperature (0-1)
+            top_p: Nucleus sampling parameter (0-1)
+            max_tokens: Maximum tokens to generate
+            **kwargs: Additional arguments passed to AgentBase
+        """
+        # Store Bedrock-specific parameters first
+        self._voice_id = voice_id
+        self._temperature = temperature
+        self._top_p = top_p
+        self._max_tokens = max_tokens
+        
+        # Initialize base class
+        super().__init__(name=name, route=route, **kwargs)
+        
+        # Set initial prompt if provided (after super init)
+        if system_prompt:
+            self.set_prompt_text(system_prompt)
+        
+        logger.info(f"BedrockAgent initialized: {name} on route {route}")
+    
+    def _render_swml(self, call_id: str = None, modifications: Optional[dict] = None) -> str:
+        """
+        Render SWML document with amazon_bedrock verb
+        
+        This method overrides the base implementation to generate
+        SWML with the amazon_bedrock verb structure that matches
+        the ai verb structure for consistency.
+        
+        Args:
+            call_id: Optional call ID for session-specific tokens
+            modifications: Optional dict of modifications to apply
+            
+        Returns:
+            SWML document as JSON string with amazon_bedrock verb
+        """
+        # Call parent to build the base SWML with ai verb
+        base_swml_json = super()._render_swml(call_id, modifications)
+        
+        # Parse the JSON to modify it
+        swml = json.loads(base_swml_json)
+        
+        # Find and transform the ai verb to amazon_bedrock
+        sections = swml.get("sections", {})
+        main_section = sections.get("main", [])
+        
+        # Look for ai verb and transform it
+        for i, verb in enumerate(main_section):
+            if "ai" in verb:
+                ai_config = verb["ai"]
+                
+                # Build amazon_bedrock verb with same structure
+                bedrock_verb = {
+                    "amazon_bedrock": {
+                        # Add voice configuration and inference params inside prompt
+                        # Note: In Bedrock, voice and inference params are part of prompt config
+                        "prompt": self._add_voice_to_prompt(ai_config.get("prompt", {})),
+                        
+                        # Copy SWAIG if present
+                        "SWAIG": ai_config.get("SWAIG", {}),
+                        
+                        # Include params only if they were explicitly set via set_params()
+                        # The C++ code ignores params for now (marked for future extensibility)
+                        "params": ai_config.get("params", {}),
+                        
+                        # Copy global_data if present
+                        "global_data": ai_config.get("global_data", {}),
+                        
+                        # Copy post_prompt if present
+                        "post_prompt": ai_config.get("post_prompt"),
+                        
+                        # Copy post_prompt_url if present
+                        "post_prompt_url": ai_config.get("post_prompt_url")
+                    }
+                }
+                
+                # Remove None values
+                bedrock_config = bedrock_verb["amazon_bedrock"]
+                bedrock_verb["amazon_bedrock"] = {
+                    k: v for k, v in bedrock_config.items() 
+                    if v is not None
+                }
+                
+                # Replace ai verb with amazon_bedrock verb
+                main_section[i] = bedrock_verb
+                break
+        
+        # Convert back to JSON string
+        return json.dumps(swml)
+    
+    def _add_voice_to_prompt(self, prompt_config: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        Add voice configuration to the prompt object
+        
+        In Bedrock, voice configuration is part of the prompt object,
+        not a separate field like in OpenAI.
+        
+        Args:
+            prompt_config: Current prompt configuration
+            
+        Returns:
+            Updated prompt configuration with voice
+        """
+        # Create a clean copy, filtering out text-model-specific parameters
+        # that don't apply to Bedrock's voice-to-voice model
+        filtered_config = {}
+        
+        # Copy over only the relevant fields
+        for key, value in prompt_config.items():
+            # Skip text-model-specific parameters
+            if key in ['barge_confidence', 'presence_penalty', 'frequency_penalty']:
+                continue
+            filtered_config[key] = value
+        
+        # Add voice_id to the prompt configuration
+        filtered_config["voice_id"] = self._voice_id
+        
+        # Add/override inference parameters (where C code expects them)
+        filtered_config["temperature"] = self._temperature
+        filtered_config["top_p"] = self._top_p
+        
+        return filtered_config
+    
+    def _build_bedrock_params(self, base_params: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        Build Bedrock-specific parameters
+        
+        Merges base parameters with Bedrock-specific inference settings.
+        
+        Args:
+            base_params: Base parameters from AgentBase
+            
+        Returns:
+            Combined parameters for Bedrock
+        """
+        # Start with base params
+        params = base_params.copy()
+        
+        # Add Bedrock inference parameters
+        params.update({
+            "temperature": self._temperature,
+            "top_p": self._top_p,
+            "max_tokens": self._max_tokens
+        })
+        
+        return params
+    
+    def set_voice(self, voice_id: str) -> None:
+        """
+        Set the Bedrock voice ID
+        
+        Args:
+            voice_id: Bedrock voice identifier (e.g., 'matthew', 'joanna')
+        """
+        self._voice_id = voice_id
+        logger.debug(f"Voice set to: {voice_id}")
+    
+    def set_inference_params(
+        self,
+        temperature: Optional[float] = None,
+        top_p: Optional[float] = None,
+        max_tokens: Optional[int] = None
+    ) -> None:
+        """
+        Update Bedrock inference parameters
+        
+        Args:
+            temperature: Generation temperature (0-1)
+            top_p: Nucleus sampling parameter (0-1)
+            max_tokens: Maximum tokens to generate
+        """
+        if temperature is not None:
+            self._temperature = temperature
+        if top_p is not None:
+            self._top_p = top_p
+        if max_tokens is not None:
+            self._max_tokens = max_tokens
+        
+        logger.debug(f"Inference params updated: temp={self._temperature}, "
+                    f"top_p={self._top_p}, max_tokens={self._max_tokens}")
+    
+    # Methods that may not be relevant to Bedrock
+    # These are overridden to provide appropriate behavior or warnings
+    
+    def set_llm_model(self, model: str) -> None:
+        """
+        Set LLM model - not applicable for Bedrock
+        
+        Bedrock uses a fixed voice-to-voice model, so this method
+        logs a warning and does nothing.
+        
+        Args:
+            model: Model name (ignored)
+        """
+        logger.warning(f"set_llm_model('{model}') called but Bedrock uses a fixed voice-to-voice model")
+    
+    def set_llm_temperature(self, temperature: float) -> None:
+        """
+        Set LLM temperature - redirects to set_inference_params
+        
+        Args:
+            temperature: Temperature value
+        """
+        self.set_inference_params(temperature=temperature)
+    
+    def set_post_prompt_llm_params(self, **params) -> None:
+        """
+        Set post-prompt LLM parameters - not applicable for Bedrock
+        
+        Bedrock uses OpenAI for post-prompt summarization, but those
+        parameters are configured in the C code.
+        
+        Args:
+            **params: Ignored parameters
+        """
+        logger.warning("set_post_prompt_llm_params() called but Bedrock post-prompt uses OpenAI configured in C code")
+    
+    def set_prompt_llm_params(self, **params) -> None:
+        """
+        Set prompt LLM parameters - use set_inference_params instead
+        
+        For Bedrock, use set_inference_params() to configure temperature,
+        top_p, and max_tokens.
+        
+        Args:
+            **params: Parameters (ignored, use set_inference_params)
+        """
+        logger.warning("set_prompt_llm_params() called - use set_inference_params() for Bedrock")
+    
+    # Note: We don't override prompt methods like set_prompt_text, set_prompt_pom
+    # because those work fine - they just build the prompt structure that we
+    # transform in _render_swml()
+    
+    def __repr__(self) -> str:
+        """String representation of the agent"""
+        return (f"BedrockAgent(name='{self.name}', route='{self.route}', "
+                f"voice='{self._voice_id}')")

signalwire_agents/cli/__init__.py

@@ -0,0 +1,18 @@
+"""
+Copyright (c) 2025 SignalWire
+
+This file is part of the SignalWire AI Agents SDK.
+
+Licensed under the MIT License.
+See LICENSE file in the project root for full license information.
+"""
+
+"""
+SignalWire Agents CLI Tools
+
+This package contains command-line tools for working with SignalWire AI Agents.
+"""
+
+from .test_swaig import main as test_swaig_main
+
+__all__ = ['test_swaig_main']