LLMem - The Lightest Memory Management Layer

A tool-independent, universal memory management system using a single bash script. LLMem provides persistent, searchable memory storage for AI agents and assistants—completely independent of any specific tool, platform, or built-in memory system.

Overview

LLMem is not about project-specific memory management—it's a lightweight, general-purpose memory layer that works anywhere. With just bash, jq, and a JSON file, you get a complete memory system that:

• Works with any AI tool or platform - No vendor lock-in
• Persists across conversations - Memories survive beyond individual sessions
• Portable and transparent - Simple JSON storage you can inspect and migrate
• Minimal dependencies - Runs wherever bash runs

Tool-Independent Design

Unlike built-in memory systems tied to specific AI platforms, LLMem is completely independent:

• Not bound to any specific AI tool, chatbot, or IDE
• No API calls, cloud services, or proprietary formats
• Works equally well with ChatGPT, Claude, Cursor, local LLMs, or your own scripts
• Simply integrate the bash script into any workflow—it's just standard I/O

This is universal memory management, not project-specific state tracking.

Storage Format

Memories are stored in llmem.json as a JSON array with the following structure:

[
  {
    "id": 1,
    "title": "Memory title",
    "content": "Detailed memory content",
    "timestamp": "2025-11-25T10:30:00"
  }
]

• id: Auto-incrementing unique identifier
• title: Short descriptive title for the memory
• content: Full memory details
• timestamp: ISO 8601 timestamp of when the memory was created

Available Commands

Save a Memory

./llmem.sh save "title" "content"

Stores a new memory with auto-generated ID and timestamp.

Search Memories

./llmem.sh search "query"

Case-insensitive search through both title and content fields using grep.

Retrieve All Memories

./llmem.sh get_all

Returns all stored memories in a formatted, readable output.

Use Cases

LLMem is a general-purpose memory layer for any context where you need persistent memory:

• AI Agent Memory - Give any AI assistant long-term memory, regardless of the platform
• Cross-Conversation Context - Maintain continuity across multiple chat sessions
• Knowledge Base - Store and retrieve insights, decisions, and learnings
• Personal Notes System - Searchable memory for your own use
• Multi-Tool Coordination - Share memory between different AI tools and platforms

Whether you're using ChatGPT, Claude, local LLMs, or custom agents—LLMem works with all of them.

Dependencies

• bash
• jq - For JSON manipulation
• grep - For searching (standard on Unix systems)

Agent Integration

See AGENTS.md for instructions on integrating this memory system with AI agents/assistants.

License

MIT License - see LICENSE file for details.