I have been using Claude Code for about ten months. The session organization story went something like this:
- Month 1: I closed a long session by accident. Lost 40 minutes of context.
- Month 3: I learned about
/resumeand/rename. Renamed every session I cared about. - Month 5: I had 60+ named sessions across 8 projects. I could not remember the names.
/resumewas useless. - Month 7: I started grepping
~/.claude/projects/*/*.jsonl. Worked, but I never knew which project folder to grep first. - Month 9: I just gave up and re-ran the work, slower.
The honest summary: Claude Code saves everything. The problem is it does not help me find anything.
I have tried the obvious things:
- terminal scrollback (gone when I close the tab)
- copying key bits to a "prompts" notes file (low fidelity)
- search by project name in
~/.claude/projects/(works but slow) - AGENTS.md and CLAUDE.md (good for future sessions, useless for past ones)
What I actually wanted: an indexed view of every session I ever ran, searchable by keyword, grouped by project, with one click to resume.
Two months ago I started using Shelf. It is a small Tauri desktop app (local, not cloud) that does exactly this: it scans ~/.claude/projects/ and ~/.codex/sessions/, shows you every session as a card, lets you search across them, and reopens the right one when you click. It also keeps a per-project view so I can see "what was I doing in repo X last week" without grepping.
Is it perfect? No. The startup scan is slow if you have a lot of sessions. The UI is functional, not beautiful. But it is the first thing I have used that actually matches the mental model I had for "find me the conversation where I solved Y".
If your Claude Code history is more than a handful of weeks deep, you have probably already felt the pain. I would be curious what other people are doing.
Tags: claudecode, codex, productivity, devtools, opensource
Top comments (0)