markmhendrickson/neotoma

What it does

Neotoma is a deterministic state layer designed for AI agents that provides a structured way to store and manage versioned entities including contacts, tasks, transactions, and decisions. It maintains immutable observations with complete provenance tracking, enabling AI systems to have reliable, auditable memory across multiple client applications. The tool implements schema-first data extraction and provides cross-client synchronization capabilities.

How to set up

Neotoma functions as an MCP (Model Context Protocol) server. Installation involves cloning the repository from GitHub and configuring it as a local-first SQLite database. The server can be integrated with Claude, Cursor, ChatGPT, and other compatible clients through MCP configuration. Detailed setup instructions are available in the GitHub repository's documentation.

Key features

• Versioned entity storage with full immutable history
• Complete provenance tracking for all observations and decisions
• Schema-first data extraction for structured information
• Local-first SQLite backend for offline-capable operation
• Cross-client memory synchronization across Claude, Cursor, ChatGPT, and compatible AI platforms
• Deterministic state management for reproducible AI agent behavior