MemoryVault-AI

MemoryVault AI is a persistent, multimodal AI memory system that allows users to store, retrieve, and chat with their own memories. It combines multimodal ingestion, persistent storage, hybrid retrieval, CrossEncoder reranking, Gemini function calling, and LangGraph into one practical AI application. The system is designed to give an AI assistant a persistent memory that it can search and use when answering questions. It supports various input types including PDF, text, image, and audio, which are converted into searchable text before entering a common RAG pipeline. The retrieval pipeline uses FAISS for semantic retrieval, BM25 for lexical retrieval, Reciprocal Rank Fusion to combine results, and CrossEncoder reranking for final relevance ordering. The agentic workflow uses Gemini function calling and LangGraph, with tool execution performed in Python. Conversations and messages are stored in SQLite, allowing persistent chat history across application restarts. The application includes a dashboard for memory statistics and exploration, and a Memory Recall capability for scheduling future reminders.