Remember Me

🚀 Remember Me

A persistent framework for maintaining conversation context and rules in MCP-based language model applications.

🚀 Quick Start

Running the Server

Using MCP Inspector

1. Clone the repository:

git clone https://github.com/yourusername/memories.git
cd memories

2. Install dependencies:

npm install

3. Start the development server:

npm run dev

Configuring the LLM Platform

Configure the LLM to communicate with the "Remember Me" server. The specific steps depend on the chosen LLM framework (e.g., OpenAI, Hugging Face, etc.).

✨ Features

"Remember Me" is an MCP server designed to persistently store and retrieve different types of resources:

• Rules: Define constraints and guidelines for interaction behavior.
• Snippets: Store reusable code or text snippets.
• Summaries: Save important context information from conversations.

This system uses SQLite for persistence and provides a comprehensive API to manage different contexts of these resources.

🔧 Technical Details

Architecture

Core Components

• MyContext: The central manager for all persistent operations.
• PersistentResource: The base class for all storable resources.
  • Rule: Commands that define interaction parameters.
  • Snippet: Code or text snippets that can be referenced.
  • Summary: Conversation context information.
• Backup: A system for creating and restoring context states.

Data Model

The storage attributes of resources are as follows:

• Context: The namespace of the resource (e.g., "me" represents global resources).
• Key: A unique identifier within the context.
• Content: The actual stored data.
• Type/MIME Type: Information for handling the format.

Rule System

Rules adopt a structured policy framework:

• MUST: Absolute requirements.
• MUST NOT: Absolute prohibitions.
• SHOULD: Recommended practices.
• SHOULD NOT: Discouraged practices.
• MAY: Optional considerations.

API

Context Management

• my_context(): Load the current context and attach additional contexts if needed.
• my_context_backup_create(): Create a backup of the current state.
• my_context_backup_restore(): Restore from a previous backup.
• my_context_backup_list(): View available backups.
• my_context_backup_remove(): Delete a specific backup.
• my_context_backup_clear(): Clear all backups.

Rule Management

• my_context_rule_list(): List all rules in the context.
• my_context_rule_set(): Create or update a rule.
• my_context_rule_remove(): Delete a rule.

Snippet Management

• my_context_snippet_list(): List snippets in the context.
• my_context_snippet_get(): Get a specific snippet.
• my_context_snippet_set(): Create or update a snippet.
• my_context_snippet_remove(): Delete a snippet.

Summary Management

• my_context_summary_list(): List summaries in the context.
• my_context_summary_get(): Get a specific summary.
• my_context_summary_set(): Create or update a summary.
• my_context_summary_remove(): Delete a summary.

Using with LLMs

The "me" Context

"me" is a special default context that is always available. It contains global rules, snippets, and summaries that apply to each conversation.

Loading Contexts

The LLM should load the context at the start of the conversation. This retrieves all rules, snippets, and summaries from the "me" context. The LLM should then follow any returned rules.

Additional Contexts

In addition to "me", you can load other contexts by specifying additional contexts. For example:

• A "coding" context that contains programming-related snippets.
• A "creative" context that contains writing prompts.
• A "technical" context that contains specialized knowledge.

These can be attached together with the default "me" context when needed.