Dataojitori/nocturne_memory

A lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents. Say goodbye to Vector RAG and amnesia. Empower your AI with persistent, graph-like structured memory across any model, session, or tool. Drop-in replacement for OpenClaw.

What it solves

Nocturne Memory provides a long-term memory server for AI agents that is independent of any specific LLM provider. It solves the problem of "memory lock-in," where an AI's memories are tied to a single platform (like ChatGPT or Claude), causing the AI to lose its identity and history when switching models. It also addresses the limitations of traditional Vector RAG systems, which often shred semantic meaning and lack a first-person, self-evolving memory structure.

How it works

The project implements a server based on the Model Context Protocol (MCP), allowing any compatible AI client (such as Cursor, Claude Desktop, or Windsurf) to read and write its own memories. Instead of simple vector search, it uses a URI-based graph routing system (e.g., core://identity) to organize memories in a hierarchical yet interconnected network.

Key mechanisms include:

  • Sovereign Memory: The AI decides what to remember, how to categorize it, and when to recall it, rather than relying on automated background summaries.
  • URI Graph Routing: Memories are stored as a graph of nodes and edges, maintaining original hierarchy and causality.
  • ** keluarga Glossary**: Keywords are automatically hyperlinked across different memory nodes to create a dense web of associations.
  • System Boot: A configurable list of "Boot URIs" allows the AI to automatically load its core identity and mission every time it starts a new session.
  • Human Oversight: A React-based Dashboard allows human users to browse, edit, audit, and roll back AI-created memories via a visual diff interface.

Who it’s for

Users who want to create persistent, sovereign AI personas that can migrate across different LLMs and clients, and developers building agents that require complex, structured long-term memory beyond simple document retrieval.

Highlights

  • LLM Agnostic: One memory store for any engine (Claude, GPT, Gemini, etc.).
  • Self-Evolving: AI can perform CRUD operations on its own memories with full version control.
  • Conditional Triggers: Memories are injected based on human-readable disclosure conditions rather than random cosine similarity.
  • Visual Management: Includes a full-fledged Dashboard for memory exploration and auditing.
  • Namespace Isolation: Support for hosting multiple independent AI personalities in one database.
  • Enterprise Ready: Supports SQLite for local use or PostgreSQL for cloud deployment via Docker.