alita-sdk 0.3.181__py3-none-any.whl → 0.3.183__py3-none-any.whl

alita_sdk/runtime/langchain/agents/react_agent.py

@@ -1,157 +0,0 @@
-import logging
-from typing import Any, Optional, Callable
-from langchain_core.messages import BaseMessage, AIMessage, HumanMessage
-from langchain_core.tools import BaseTool
-from langgraph.prebuilt import create_react_agent
-from langgraph.checkpoint.memory import MemorySaver
-
-logger = logging.getLogger(__name__)
-
-def create_react_agent_with_auto_continue(
-    prompt: Any,
-    model: Any,
-    tools: list[BaseTool],
-    memory: Optional[Any] = None,
-    debug: bool = False
-) -> Any:
-    """
-    Create a LangGraph React agent with auto-continue capability for when responses get truncated.
-    This provides better handling of length-limited responses compared to traditional AgentExecutor.
-    Uses simple in-memory checkpointing for auto-continue functionality.
-    
-    Args:
-        prompt: The prompt template to use for the agent
-        model: The language model to use
-        tools: List of tools available to the agent
-        memory: Optional memory store for checkpointing (will create MemorySaver if None)
-        debug: Whether to enable debug mode
-        
-    Returns:
-        A configured React agent with auto-continue capability
-        
-    Note: Requires LangGraph 0.5.x or higher that supports post_model_hook.
-    """
-    # Use simple in-memory checkpointer for auto-continue functionality if not provided
-    if memory is None:
-        memory = MemorySaver()
-    
-    # Set up parameters for the agent
-    kwargs = {
-        "prompt": prompt,
-        "model": model,
-        "tools": tools,
-        "checkpointer": memory,
-        "post_model_hook": _create_auto_continue_hook()  # Auto-continue hook
-    }
-
-    # Create the base React agent with langgraph's prebuilt function
-    base_agent = create_react_agent(**kwargs)
-    
-    return base_agent
-
-def _create_auto_continue_hook() -> Callable:
-    """
-    Create a post-model hook for LangGraph 0.5.x that detects truncated responses
-    and adds continuation prompts.
-    This checks if the last AI message was truncated and automatically continues if needed.
-    """
-    MAX_CONTINUATIONS = 3  # Maximum number of auto-continuations allowed
-    
-    def post_model_hook(state):
-        messages = state.get("messages", [])
-        
-        # Count how many auto-continue messages we've already sent
-        continuation_count = sum(
-            1 for msg in messages 
-            if isinstance(msg, HumanMessage) and 
-            "continue your previous response" in msg.content.lower()
-        )
-        
-        # Don't continue if we've reached the limit
-        if continuation_count >= MAX_CONTINUATIONS:
-            return state
-            
-        # Check if the last message is from AI and was truncated
-        if messages and isinstance(messages[-1], AIMessage):
-            last_ai_message = messages[-1]
-            
-            # Check for truncation indicators
-            is_truncated = (
-                hasattr(last_ai_message, 'response_metadata') and 
-                last_ai_message.response_metadata.get('finish_reason') == 'length'
-            ) or (
-                # Fallback: check if message seems to end abruptly
-                last_ai_message.content and 
-                not last_ai_message.content.rstrip().endswith(('.', '!', '?', ':', ';'))
-            )
-            
-            # Add continuation request if truncated
-            if is_truncated:
-                logger.info("Detected truncated response, adding continuation request")
-                new_messages = messages.copy()
-                new_messages.append(HumanMessage(content="Continue your previous response from where you left off"))
-                return {"messages": new_messages}
-        
-        return state
-        
-    return post_model_hook
-
-def get_langgraph_agent_with_auto_continue(
-    prompt: Any,
-    model: Any,
-    tools: list[BaseTool],
-    memory: Optional[Any] = None,
-    debug: bool = False
-) -> Any:
-    """
-    Create a LangGraph agent with auto-continue capability for when responses get truncated.
-    This provides better handling of length-limited responses compared to traditional AgentExecutor.
-    Uses simple in-memory checkpointing for auto-continue functionality.
-    
-    Args:
-        prompt: The prompt template to use for the agent
-        model: The language model to use
-        tools: List of tools available to the agent
-        memory: Optional memory store for checkpointing (will create MemorySaver if None)
-        debug: Whether to enable debug mode
-        
-    Returns:
-        A configured LangGraphAgentRunnable with auto-continue capability
-        
-    Note: Requires LangGraph 0.5.x or higher that supports post_model_hook.
-    """
-    from ...langchain.langraph_agent import LangGraphAgentRunnable
-    
-    # Use simple in-memory checkpointer for auto-continue functionality if not provided
-    if memory is None:
-        memory = MemorySaver()
-    
-    # Create the base React agent with auto-continue capability
-    base_agent = create_react_agent_with_auto_continue(
-        prompt=prompt,
-        model=model,
-        tools=tools,
-        memory=memory,
-        debug=debug
-    )
-    
-    # Wrap the base agent in our custom LangGraphAgentRunnable to handle input properly
-    # This ensures that our invoke() input handling logic is applied
-    agent = LangGraphAgentRunnable(
-        builder=base_agent.builder,
-        config_type=base_agent.builder.config_schema,
-        nodes=base_agent.nodes,
-        channels=base_agent.channels,
-        input_channels=base_agent.input_channels,
-        stream_mode=base_agent.stream_mode,
-        output_channels=base_agent.output_channels,
-        stream_channels=base_agent.stream_channels,
-        checkpointer=memory,
-        interrupt_before_nodes=base_agent.interrupt_before_nodes,
-        interrupt_after_nodes=base_agent.interrupt_after_nodes,
-        debug=debug,
-        store=base_agent.store, 
-        schema_to_mapper=base_agent.schema_to_mapper
-    )
-    
-    return agent