Claude Recap

Claude-Recap

  Topic-based automatic memory for Claude Code — never lose context across sessions or compactions.

  English | 中文

The Problem

Claude Code forgets everything between sessions. Switch topics mid-conversation and the previous context is gone. Hit a context compaction and your working state evaporates. Start a new session and you're explaining the same project from scratch.

What Claude-Recap Does

Two shell hooks that run automatically — zero manual effort:

- Automatic topic archival — Every response gets a topic tag. When the topic changes, the previous one is summarized and saved to a Markdown file.
- Context injection — Each new session starts with your topic history and remembered preferences injected automatically.
- Compaction recovery — When Claude Code compacts your context, Claude-Recap cold-reads from the JSONL transcript to rebuild accurate summaries. Nothing is lost.
- /remember skill — Tell Claude to remember preferences across sessions: "always use bun", "never auto-commit". Stored in plain Markdown.

Everything is stored locally as Markdown files in ~/.memory/. No database, no cloud, no dependencies beyond bash and Node.js.

Quick Start

Plugin install (recommended)
Note: After install, restart Claude Code for hooks to take effect. /remember works immediately, but topic features (/save-topic, /list-topics, auto-archival) require the restart to inject the Topic Tag Rule.

Manual install (without plugin system)

This writes hook entries directly into your project's .claude/settings.json.

How It Works

Features

Storage

All data lives in ~/.memory/ (configurable via MEMORY_HOME env var):

Comparison

Configuration

Uninstall

Your data in ~/.memory/ is preserved — uninstalling does not delete memory files. Reinstalling restores full…