Lancedb Memory

Manage and retrieve long-term memories with LanceDB using semantic vector search, category filtering, and detailed metadata storage.

#!/usr/bin/env python3 """ LanceDB integration for long-term memory management. Provides vector search and semantic memory capabilities. """

import os import json import lancedb from datetime import datetime from typing import List, Dict, Any, Optional from pathlib import Path

class LanceMemoryDB: """LanceDB wrapper for long-term memory storage and retrieval."""

text

def __init__(self, db_path: str = "/Users/prerak/clawd/memory/lancedb"):
    self.db_path = Path(db_path)
    self.db_path.mkdir(parents=True, exist_ok=True)
    self.db = lancedb.connect(self.db_path)
    
    # Ensure memory table exists
    if "memory" not in self.db.table_names():
        self._create_memory_table()

def _create_memory_table(self):
    """Create the memory table with appropriate schema."""
    schema = [
        {"name": "id", "type": "int", "nullable": False},
        {"name": "timestamp", "type": "timestamp", "nullable": False},
        {"name": "content", "type": "str", "nullable": False},
        {"name": "category", "type": "str", "nullable": True},
        {"name": "tags", "type": "str[]", "nullable": True},
        {"name": "importance", "type": "int", "nullable": True},
        {"name": "metadata", "type": "json", "nullable": True},
    ]
    
    self.db.create_table("memory", schema=schema)

def add_memory(self, content: str, category: str = "general", tags: List[str] = None, 
               importance: int = 5, metadata: Dict[str, Any] = None) -> int:
    """Add a new memory entry."""
    table = self.db.open_table("memory")
    
    # Get next ID
    max_id = table.to_pandas()["id"].max() if len(table) > 0 else 0
    new_id = max_id + 1
    
    # Insert new memory
    memory_data = {
        "id": new_id,
        "timestamp": datetime.now(),
        "content": content,
        "category": category,
        "tags": tags or [],
        "importance": importance,
        "metadata": metadata or {}
    }
    
    table.add([memory_data])
    return new_id

def search_memories(self, query: str, category: str = None, limit: int = 10) -> List[Dict]:
    """Search memories using vector similarity."""
    table = self.db.open_table("memory")
    
    # Build filter
    where_clause = []
    if category:
        where_clause.append(f"category = '{category}'")
    
    filter_expr = " AND ".join(where_clause) if where_clause else None
    
    # Vector search
    results = table.vector_search(query).limit(limit).where(filter_expr).to_list()
    
    return results

def get_memories_by_category(self, category: str, limit: int = 50) -> List[Dict]:
    """Get memories by category."""
    table = self.db.open_table("memory")
    df = table.to_pandas()
    filtered = df[df["category"] == category].head(limit)
    return filtered.to_dict("records")

def get_memory_by_id(self, memory_id: int) -> Optional[Dict]:
    """Get a specific memory by ID."""
    table = self.db.open_table("memory")
    df = table.to_pandas()
    result = df[df["id"] == memory_id]
    return result.to_dict("records")[0] if len(result) > 0 else None

def update_memory(self, memory_id: int, **kwargs) -> bool:
    """Update a memory entry."""
    table = self.db.open_table("memory")
    
    valid_fields = ["content", "category", "tags", "importance", "metadata"]
    updates = {k: v for k, v in kwargs.items() if k in valid_fields}
    
    if not updates:
        return False
    
    # Convert to proper types for LanceDB
    if "tags" in updates and isinstance(updates["tags"], list):
        updates["tags"] = str(updates["tags"]).replace("'", '"')
    
    table.update(updates, where=f"id = {memory_id}")
    return True

def delete_memory(self, memory_id: int) -> bool:
    """Delete a memory entry."""
    table = self.db.open_table("memory")
    current_count = len(table)
    table.delete(f"id = {memory_id}")
    return len(table) < current_count

def get_all_categories(self) -> List[str]:
    """Get all unique categories."""
    table = self.db.open_table("memory")
    df = table.to_pandas()
    return df["category"].dropna().unique().tolist()

def get_memory_stats(self) -> Dict[str, Any]:
    """Get statistics about memory storage."""
    table = self.db.open_table("memory")
    df = table.to_pandas()
    
    return {
        "total_memories": len(df),
        "categories": len(self.get_all_categories()),
        "by_category": df["category"].value_counts().to_dict(),
        "date_range": {
            "earliest": df["timestamp"].min().isoformat() if len(df) > 0 else None,
            "latest": df["timestamp"].max().isoformat() if len(df) > 0 else None
        }
    }

Global instance

lancedb_memory = LanceMemoryDB()

def add_memory(content: str, category: str = "general", tags: List[str] = None, importance: int = 5, metadata: Dict[str, Any] = None) -> int: """Add a memory to the LanceDB store.""" return lancedb_memory.add_memory(content, category, tags, importance, metadata)

def search_memories(query: str, category: str = None, limit: int = 10) -> List[Dict]: """Search memories using semantic similarity.""" return lancedb_memory.search_memories(query, category, limit)

def get_memories_by_category(category: str, limit: int = 50) -> List[Dict]: """Get memories by category.""" return lancedb_memory.get_memories_by_category(category, limit)

def get_memory_stats() -> Dict[str, Any]: """Get memory storage statistics.""" return lancedb_memory.get_memory_stats()

Example usage

if name == "main": # Test the database print("Testing LanceDB memory integration...")

text

# Add a test memory
test_id = add_memory(
    content="This is a test memory for LanceDB integration",
    category="test",
    tags=["lancedb", "integration", "test"],
    importance=8
)
print(f"Added memory with ID: {test_id}")

# Search for memories
results = search_memories("test memory")
print(f"Search results: {len(results)} memories found")

# Get stats
stats = get_memory_stats()
print(f"Memory stats: {stats}")