""" PsycheOS — FAISS Memory Store Cross-session RAG for patient history retrieval. """ import os import json import hashlib from typing import List, Optional from datetime import datetime try: import numpy as np import faiss from sentence_transformers import SentenceTransformer FAISS_AVAILABLE = True except ImportError: FAISS_AVAILABLE = False class MemoryStore: """ Stores and retrieves conversation history using FAISS vector search. Falls back to keyword-based retrieval if FAISS/sentence-transformers unavailable. """ def __init__(self, dim: int = 384, index_path: str = "/tmp/psycheos_faiss"): self.dim = dim self.index_path = index_path self.memories = [] # list of {text, distress, timestamp, embedding_idx} self.faiss_available = FAISS_AVAILABLE self._encoder = None self._index = None if self.faiss_available: try: self._encoder = SentenceTransformer("all-MiniLM-L6-v2") self._index = faiss.IndexFlatL2(dim) except Exception: self.faiss_available = False def store(self, message: str, distress_level: int, session: int = 1): """Store a message with its distress level and metadata.""" entry = { "text": message, "distress": distress_level, "session": session, "timestamp": datetime.now().isoformat(), } if self.faiss_available and self._encoder and self._index: try: emb = self._encoder.encode([message]) self._index.add(emb) entry["embedding_idx"] = self._index.ntotal - 1 except Exception: pass self.memories.append(entry) def retrieve(self, query: str, k: int = 3) -> str: """Retrieve top-k relevant past memories.""" if not self.memories: return "" if self.faiss_available and self._encoder and self._index and self._index.ntotal > 0: try: import numpy as np q_emb = self._encoder.encode([query]) k_actual = min(k, self._index.ntotal) distances, indices = self._index.search(q_emb, k_actual) retrieved = [] for idx in indices[0]: if 0 <= idx < len(self.memories): m = self.memories[idx] retrieved.append(f"[Distress {m['distress']}] {m['text'][:80]}") return " | ".join(retrieved) except Exception: pass # Keyword fallback query_words = set(query.lower().split()) scored = [] for m in self.memories[-20:]: mem_words = set(m["text"].lower().split()) overlap = len(query_words & mem_words) scored.append((overlap, m)) scored.sort(reverse=True) top = scored[:k] if not top: return "" return " | ".join( f"[Distress {m['distress']}] {m['text'][:60]}" for _, m in top if m["text"] ) def get_session_summary(self) -> dict: """Return a summary of stored memories.""" if not self.memories: return {"total": 0, "avg_distress": 0, "max_distress": 0} distress_levels = [m["distress"] for m in self.memories] return { "total": len(self.memories), "avg_distress": round(sum(distress_levels) / len(distress_levels), 2), "max_distress": max(distress_levels), }