SimpleMem/MCP/reference/main.py

"""
SimpleMem - Efficient Lifelong Memory for LLM Agents
Main system class integrating all components
"""
from typing import List, Optional
from models.memory_entry import Dialogue, MemoryEntry
from utils.llm_client import LLMClient
from utils.embedding import EmbeddingModel
from database.vector_store import VectorStore
from core.memory_builder import MemoryBuilder
from core.hybrid_retriever import HybridRetriever
from core.answer_generator import AnswerGenerator
import config


class SimpleMemSystem:
    """
    SimpleMem Main System

    Three-stage pipeline based on Semantic Lossless Compression:
    1. Semantic Structured Compression: add_dialogue() -> MemoryBuilder -> VectorStore
    2. Structured Indexing and Recursive Consolidation: (background evolution - future work)
    3. Adaptive Query-Aware Retrieval: ask() -> HybridRetriever -> AnswerGenerator
    """
    def __init__(
        self,
        api_key: Optional[str] = None,
        model: Optional[str] = None,
        base_url: Optional[str] = None,
        db_path: Optional[str] = None,
        table_name: Optional[str] = None,
        clear_db: bool = False,
        enable_thinking: Optional[bool] = None,
        use_streaming: Optional[bool] = None,
        enable_planning: Optional[bool] = None,
        enable_reflection: Optional[bool] = None,
        max_reflection_rounds: Optional[int] = None,
        enable_parallel_processing: Optional[bool] = None,
        max_parallel_workers: Optional[int] = None,
        enable_parallel_retrieval: Optional[bool] = None,
        max_retrieval_workers: Optional[int] = None
    ):
        """
        Initialize system

        Args:
        - api_key: OpenAI API key
        - model: LLM model name
        - base_url: Custom OpenAI base URL (for compatible APIs)
        - db_path: Database path
        - table_name: Memory table name (for parallel processing)
        - clear_db: Whether to clear existing database
        - enable_thinking: Enable deep thinking mode (for Qwen and compatible models)
        - use_streaming: Enable streaming responses
        - enable_planning: Enable multi-query planning for retrieval (None=use config default)
        - enable_reflection: Enable reflection-based additional retrieval (None=use config default)
        - max_reflection_rounds: Maximum number of reflection rounds (None=use config default)
        - enable_parallel_processing: Enable parallel processing for memory building (None=use config default)
        - max_parallel_workers: Maximum number of parallel workers for memory building (None=use config default)
        - enable_parallel_retrieval: Enable parallel processing for retrieval queries (None=use config default)
        - max_retrieval_workers: Maximum number of parallel workers for retrieval (None=use config default)
        """
        print("=" * 60)
        print("Initializing SimpleMem System")
        print("=" * 60)

        # Initialize core components
        self.llm_client = LLMClient(
            api_key=api_key,
            model=model,
            base_url=base_url,
            enable_thinking=enable_thinking,
            use_streaming=use_streaming
        )
        self.embedding_model = EmbeddingModel()
        self.vector_store = VectorStore(
            db_path=db_path,
            embedding_model=self.embedding_model,
            table_name=table_name
        )

        if clear_db:
            print("\nClearing existing database...")
            self.vector_store.clear()

        # Initialize three major modules
        self.memory_builder = MemoryBuilder(
            llm_client=self.llm_client,
            vector_store=self.vector_store,
            enable_parallel_processing=enable_parallel_processing,
            max_parallel_workers=max_parallel_workers
        )

        self.hybrid_retriever = HybridRetriever(
            llm_client=self.llm_client,
            vector_store=self.vector_store,
            enable_planning=enable_planning,
            enable_reflection=enable_reflection,
            max_reflection_rounds=max_reflection_rounds,
            enable_parallel_retrieval=enable_parallel_retrieval,
            max_retrieval_workers=max_retrieval_workers
        )

        self.answer_generator = AnswerGenerator(
            llm_client=self.llm_client
        )

        print("\nSystem initialization complete!")
        print("=" * 60)

    def add_dialogue(self, speaker: str, content: str, timestamp: Optional[str] = None):
        """
        Add a single dialogue

        Args:
        - speaker: Speaker name
        - content: Dialogue content
        - timestamp: Timestamp (ISO 8601 format)
        """
        dialogue_id = self.memory_builder.processed_count + len(self.memory_builder.dialogue_buffer) + 1
        dialogue = Dialogue(
            dialogue_id=dialogue_id,
            speaker=speaker,
            content=content,
            timestamp=timestamp
        )
        self.memory_builder.add_dialogue(dialogue)

    def add_dialogues(self, dialogues: List[Dialogue]):
        """
        Batch add dialogues

        Args:
        - dialogues: List of dialogues
        """
        self.memory_builder.add_dialogues(dialogues)

    def finalize(self):
        """
        Finalize dialogue input, process any remaining buffer (safety check)
        Note: In parallel mode, remaining dialogues are already processed
        """
        self.memory_builder.process_remaining()

    def ask(self, question: str) -> str:
        """
        Ask question - Core Q&A interface

        Args:
        - question: User question

        Returns:
        - Answer
        """
        print("\n" + "=" * 60)
        print(f"Question: {question}")
        print("=" * 60)

        # Stage 2: Hybrid retrieval
        contexts = self.hybrid_retriever.retrieve(question)

        # Stage 3: Answer generation
        answer = self.answer_generator.generate_answer(question, contexts)

        print("\nAnswer:")
        print(answer)
        print("=" * 60 + "\n")

        return answer

    def get_all_memories(self) -> List[MemoryEntry]:
        """
        Get all memory entries (for debugging)
        """
        return self.vector_store.get_all_entries()

    def print_memories(self):
        """
        Print all memory entries (for debugging)
        """
        memories = self.get_all_memories()
        print("\n" + "=" * 60)
        print(f"All Memory Entries ({len(memories)} total)")
        print("=" * 60)

        for i, memory in enumerate(memories, 1):
            print(f"\n[Entry {i}]")
            print(f"ID: {memory.entry_id}")
            print(f"Restatement: {memory.lossless_restatement}")
            if memory.timestamp:
                print(f"Time: {memory.timestamp}")
            if memory.location:
                print(f"Location: {memory.location}")
            if memory.persons:
                print(f"Persons: {', '.join(memory.persons)}")
            if memory.entities:
                print(f"Entities: {', '.join(memory.entities)}")
            if memory.topic:
                print(f"Topic: {memory.topic}")
            print(f"Keywords: {', '.join(memory.keywords)}")

        print("\n" + "=" * 60)


# Convenience function
def create_system(
    clear_db: bool = False,
    enable_planning: Optional[bool] = None,
    enable_reflection: Optional[bool] = None,
    max_reflection_rounds: Optional[int] = None,
    enable_parallel_processing: Optional[bool] = None,
    max_parallel_workers: Optional[int] = None,
    enable_parallel_retrieval: Optional[bool] = None,
    max_retrieval_workers: Optional[int] = None
) -> SimpleMemSystem:
    """
    Create SimpleMem system instance (uses config.py defaults when None)
    """
    return SimpleMemSystem(
        clear_db=clear_db,
        enable_planning=enable_planning,
        enable_reflection=enable_reflection,
        max_reflection_rounds=max_reflection_rounds,
        enable_parallel_processing=enable_parallel_processing,
        max_parallel_workers=max_parallel_workers,
        enable_parallel_retrieval=enable_parallel_retrieval,
        max_retrieval_workers=max_retrieval_workers
    )


if __name__ == "__main__":
    # Quick test with Qwen3 integration
    print("🚀 Running SimpleMem Quick Test with Qwen3...")

    system = create_system(clear_db=True)
    print(f"📌 Using embedding model: {system.memory_builder.vector_store.embedding_model.model_name}")
    print(f"📌 Model type: {system.memory_builder.vector_store.embedding_model.model_type}")

    # Add some test dialogues
    system.add_dialogue("Alice", "Bob, let's meet at Starbucks tomorrow at 2pm to discuss the new product", "2025-11-15T14:30:00")
    system.add_dialogue("Bob", "Okay, I'll prepare the materials", "2025-11-15T14:31:00")
    system.add_dialogue("Alice", "Remember to bring the market research report from last time", "2025-11-15T14:32:00")

    # Finalize input
    system.finalize()

    # View memories
    system.print_memories()

    # Ask questions (with new features)
    print("\n🔍 Testing retrieval with planning and reflection...")
    system.ask("When will Alice and Bob meet?")
    
    print("\n🔍 Testing adversarial question (reflection disabled)...")
    question = "What is Alice's favorite food?"
    contexts = system.hybrid_retriever.retrieve(question, enable_reflection=False)
    answer = system.answer_generator.generate_answer(question, contexts)
    print(f"\nQuestion: {question}")
    print(f"Answer: {answer}")
    
    print("\n✅ Quick test completed!")
    print("\n💡 To run comprehensive tests: python test_qwen3_integration.py")