Claude Rework

claude-rework

[](https://pypi.org/project/claude-rework/) [](https://github.com/Luneswan/claude-rework/actions/workflows/ci.yml) [](LICENSE)

Claude starts every session with no memory of the last one. Everything you decided, tried, and ruled out is sitting in `~/.claude/projects` as raw JSON that nothing reads. Ask about it and Claude either guesses or re-reads a few hundred megabytes to find out.

claude-rework indexes those transcripts once and answers from the index. A question about your own past work costs about 2,500 tokens and takes under a second, instead of 211,000 tokens and a long wait.

pip install claude-rework && claude-rework install

After that you do not type anything. Ask Claude a question about the past in plain English and the answer is already in front of it.

---

Contents

• What it saves
• Install
• How it works
• Asking it things
• Moving to a new machine or account
• Saving facts yourself
• Keeping it working
• Every command
• Privacy
• How it compares
• Requirements
• Limitations

---

What it saves

One question about your past work, measured on a real installation of 1,421 transcript files totalling 7.4 GB:

Answering "what did we decide about X?"	Input tokens
grep your notes and transcripts	211,627
claude-rework	2,552

That is **209,075 tokens saved per question**. Resuming a session works the same way: re-reading a conversation to find where you left off costs 30,000–80,000 tokens, while `--brief` rebuilds the same picture from the index for about 600, so **4