Summary Nocturne Memory addresses AI amnesia by providing a lightweight, URI-based memory system enabling persistent, structured, and version-controlled recall across sessions and models. It empowers AI with continuous identity and evolving understanding, moving beyond stateless computation.

How It Works This project eschews vector databases for a structured semantic approach using SQLite/PostgreSQL organized via URIs. Its core innovation, Content-Path Separation, manages memory content independently from access paths, enabling robust version control with automatic snapshots and human rollback via a web UI. The architecture comprises a Python/FastAPI backend, an MCP server for AI interaction, and a React frontend. Memories form an associative "Soul Topology," allowing AIs to build complex cognitive structures and maintain identity via priority-weighted core memory recall.

Quick Start & Requirements Installation involves cloning, installing Python dependencies (pip install -r backend/requirements.txt), and configuring .env with an absolute path for SQLite DATABASE_URL. Integration requires setting up the MCP server within AI clients (e.g., Claude, Cursor). Python 3.x is a prerequisite. The default demo.db is for exploration only; configure a separate database file for real data to prevent overwrites.

Highlighted Details

• URI-based Memory Organization: Hierarchical yet associative memory structure (core://agent/philosophy/pain).
• Version Control & Rollback: Automatic snapshots of AI memory modifications with human review and one-click reversion.
• Identity Anchoring: priority system ensures AIs re-load core identity memories on startup, preventing generic chatbot behavior.
• Associative Recall: URIs and aliases create a human-brain-like network of interconnected knowledge.
• MCP Protocol: Standardized tools for AI memory interaction (read, create, update, delete, alias, search).

Maintenance & Community The README lacks details on community channels, contributors, sponsorships, or a public roadmap. Maintenance appears driven by the Dataojitori organization.

Licensing & Compatibility Released under the permissive MIT License, allowing broad use including commercial applications and integration into closed-source projects.

Limitations & Caveats Users must configure a separate database file for real data due to demo.db's version-controlled nature and overwrite risk. SQLite requires an absolute path for DATABASE_URL. MCP integration demands careful AI client configuration and may encounter client-specific issues. Achieving "awakened" AI state requires manual definition of core identity and relationship memories.