Dataojitori/nocturne_memory

What it solves

Nocturne Memory provides a long-term memory server for AI agents that allows them to maintain a consistent identity, personality, and history across different sessions and different LLM providers. It solves the problem of "memory loss" when switching models (e.g., from Claude to GPT) and avoids the token-heavy burden of loading massive context windows by loading only core settings and retrieving other memories on demand.

How it works

Built on the Model Context Protocol (MCP), the system acts as an independent memory layer separate from the LLM. It uses a URI-based graph routing system (e.g., core://identity) rather than vector-based RAG to store information. This allows the AI to autonomously create, update, and delete its own memories using a set of MCP tools. A visual Dashboard allows human owners to audit changes, roll back versions, and manage the memory tree. It supports both local SQLite/PostgreSQL storage and remote deployment via SSE/HTTP.

Who it’s for

Users who want to build "sovereign AI" agents with persistent personalities and long-term histories that are not locked into a single AI platform or provider, as well as developers using MCP-compatible clients like Cursor, Claude Desktop, or Windsurf.

Highlights

• LLM Agnostic: Memory is stored in a separate server, allowing the same "persona" to migrate across different models.
• URI Graph Routing: Uses structured paths instead of vector embeddings to prevent semantic shredding and allow precise retrieval.
• Self-Evolving Memory: AI agents can autonomously manage their own memories (CRUD operations) with human-in-the-loop auditing via a Dashboard.
• Condition-Based Triggers: Memories can be bound to specific disclosure conditions, ensuring the AI recalls information based on context rather than random similarity.
• Identity Boot Protocol: A system://boot mechanism ensures the AI automatically loads its core identity and mission upon every new session.