|
| 1 | +""" |
| 2 | +Persistent Memory Bank for Plexir using ChromaDB. |
| 3 | +""" |
| 4 | + |
| 5 | +import os |
| 6 | +import logging |
| 7 | +import uuid |
| 8 | +from typing import List, Dict, Any, Optional |
| 9 | + |
| 10 | +try: |
| 11 | + import chromadb |
| 12 | + from chromadb.config import Settings |
| 13 | + from sentence_transformers import SentenceTransformer |
| 14 | + HAS_MEMORY_DEPS = True |
| 15 | +except ImportError: |
| 16 | + HAS_MEMORY_DEPS = False |
| 17 | + |
| 18 | +logger = logging.getLogger(__name__) |
| 19 | + |
| 20 | +MEMORY_DIR = os.path.expanduser("~/.plexir/memory") |
| 21 | + |
| 22 | +class MemoryBank: |
| 23 | + _instance = None |
| 24 | + |
| 25 | + def __new__(cls): |
| 26 | + if cls._instance is None: |
| 27 | + cls._instance = super(MemoryBank, cls).__new__(cls) |
| 28 | + cls._instance.initialized = False |
| 29 | + return cls._instance |
| 30 | + |
| 31 | + def __init__(self): |
| 32 | + if self.initialized: |
| 33 | + return |
| 34 | + |
| 35 | + if not HAS_MEMORY_DEPS: |
| 36 | + logger.warning("MemoryBank dependencies (chromadb, sentence-transformers) not found. Memory features disabled.") |
| 37 | + self.initialized = False |
| 38 | + return |
| 39 | + |
| 40 | + os.makedirs(MEMORY_DIR, exist_ok=True) |
| 41 | + |
| 42 | + try: |
| 43 | + self.client = chromadb.PersistentClient(path=MEMORY_DIR) |
| 44 | + |
| 45 | + # Use a lightweight model for local embeddings |
| 46 | + self.embedder = SentenceTransformer('all-MiniLM-L6-v2') |
| 47 | + |
| 48 | + self.collection = self.client.get_or_create_collection( |
| 49 | + name="plexir_memory", |
| 50 | + metadata={"hnsw:space": "cosine"} |
| 51 | + ) |
| 52 | + self.initialized = True |
| 53 | + logger.info("MemoryBank initialized with ChromaDB.") |
| 54 | + except Exception as e: |
| 55 | + logger.error(f"Failed to initialize MemoryBank: {e}") |
| 56 | + self.initialized = False |
| 57 | + |
| 58 | + def add(self, text: str, metadata: Dict[str, Any] = None) -> str: |
| 59 | + if not self.initialized: |
| 60 | + return "MemoryBank not initialized." |
| 61 | + |
| 62 | + try: |
| 63 | + doc_id = str(uuid.uuid4()) |
| 64 | + embedding = self.embedder.encode(text).tolist() |
| 65 | + |
| 66 | + self.collection.add( |
| 67 | + documents=[text], |
| 68 | + embeddings=[embedding], |
| 69 | + metadatas=[metadata or {}], |
| 70 | + ids=[doc_id] |
| 71 | + ) |
| 72 | + return f"Memory saved (ID: {doc_id})" |
| 73 | + except Exception as e: |
| 74 | + logger.error(f"Failed to add memory: {e}") |
| 75 | + return f"Error saving memory: {e}" |
| 76 | + |
| 77 | + def search(self, query: str, n_results: int = 5) -> List[Dict[str, Any]]: |
| 78 | + if not self.initialized: |
| 79 | + return [] |
| 80 | + |
| 81 | + try: |
| 82 | + query_embedding = self.embedder.encode(query).tolist() |
| 83 | + |
| 84 | + results = self.collection.query( |
| 85 | + query_embeddings=[query_embedding], |
| 86 | + n_results=n_results |
| 87 | + ) |
| 88 | + |
| 89 | + # Flatten results structure |
| 90 | + documents = results['documents'][0] |
| 91 | + metadatas = results['metadatas'][0] |
| 92 | + ids = results['ids'][0] |
| 93 | + distances = results['distances'][0] |
| 94 | + |
| 95 | + formatted_results = [] |
| 96 | + for i in range(len(documents)): |
| 97 | + formatted_results.append({ |
| 98 | + "id": ids[i], |
| 99 | + "content": documents[i], |
| 100 | + "metadata": metadatas[i], |
| 101 | + "score": 1 - distances[i] # Convert distance to similarity score |
| 102 | + }) |
| 103 | + |
| 104 | + return formatted_results |
| 105 | + except Exception as e: |
| 106 | + logger.error(f"Memory search failed: {e}") |
| 107 | + return [] |
| 108 | + |
| 109 | + def delete(self, doc_id: str) -> str: |
| 110 | + if not self.initialized: |
| 111 | + return "MemoryBank not initialized." |
| 112 | + try: |
| 113 | + self.collection.delete(ids=[doc_id]) |
| 114 | + return f"Memory {doc_id} deleted." |
| 115 | + except Exception as e: |
| 116 | + return f"Error deleting memory: {e}" |
0 commit comments