easy-agent-mem 0.3.0

Persistent memory layer for AI coding agents (Cursor / VS Code). Auto-summarizes sessions to Markdown / Obsidian.

agent-mem

Persistent memory layer for AI coding agents (Cursor, VS Code + Claude Code, etc.)

Tired of repeating context every time you start a new chat? agent-mem automatically maintains a clean project memory file so your agent always knows what happened before - without bloating the context window.

Features

• Simple local fallback: .agent-memory/memory.md
• Optional Obsidian vault support (with full graph view, backlinks, Canvas)
• Auto-generates strong agent instruction rules
• Zero extra models or API keys needed
• Works with any MCP-compatible IDE (Cursor, VS Code, etc.)
• Extremely lightweight

Installation

pip install easy-agent-mem

Quick Start

# 1. One-time setup
agent-mem init

# 2. (Recommended) Add the generated rules to your IDE
#    - Cursor: Settings -> Custom Instructions
#    - VS Code: Create CLAUDE.md or .claude/instructions.md in project root

During init you can:

• Provide an Obsidian vault path (optional)
• Or just press Enter to use the simple local .agent-memory/memory.md fallback

How It Works

1. agent-mem init creates:

   • AGENT-MEM-RULES.md (strong instructions for the agent)
   • .agent-memory/memory.md (or Obsidian notes)

2. Add the rules from AGENT-MEM-RULES.md to your IDE's custom instructions.

3. From then on, your agent will:

   • Read memory first in every new chat
   • Summarize sessions when context gets long
   • Keep a clean, persistent project history

Example Usage in Chat

Tell your agent:

"Summarize this session for memory"

It will create a clean summary and append it to memory. Then start a fresh chat - the agent will automatically load the latest memory.

Commands

agent-mem init          # Setup (Obsidian optional)
agent-mem status        # Show current config
agent-mem --help        # Full help

Project Links

• PyPI: https://pypi.org/project/easy-agent-mem/
• Source: https://github.com/atharvavdeo/agent-mem

License

MIT