Claude Never Forgets

A Claude Code plugin that provides persistent memory across sessions and context windows.

Why

Claude Code's context window is limited to 200K tokens. During long coding sessions, earlier parts of your conversation get summarized or dropped. When you start a new session, Claude has no memory of your previous work.

This plugin saves important parts of your conversations to a local file and automatically loads them when you start a new session - so your preferences, decisions, and project context persist.

Install

/plugin install yldrmahmet/claude-never-forgets

How It Works

Automatic saving:

• Every user message and Claude response is saved
• Tool rejections (when you decline a suggested action) are captured as corrections
• Memories are stored in .claude/memories/project_memory.json

Automatic loading:

• When you start a new session, saved memories are loaded into context
• Claude sees your previous preferences and decisions

Automatic cleanup:

• When memories exceed 10 entries, Claude consolidates them
• Keeps important information (preferences, decisions, corrections)
• Removes noise (greetings, acknowledgments)
• The threshold (10) can be changed in hooks/stop_cleanup.py

Commands

/remember [text]   # Manually add something to memory
/forget [text]     # Remove a specific memory
/memories          # View all stored memories

Example

Session 1:
You: "Use pnpm instead of npm for this project"
Claude: *uses pnpm*
→ Saved to memory

Session 2 (next day):
🧠 Loaded 3 memories
You: "Install axios"
Claude: *runs pnpm add axios*
→ Remembered your preference

Storage

Each project has its own memory file:

your-project/
└── .claude/
    └── memories/
        └── project_memory.json

You can:

• Commit it to git to share with your team
• Add it to .gitignore for personal preferences
• Edit it manually if needed

Privacy

All memories are stored locally. Nothing is sent to external servers.

License

MIT