🗂️ Claude Session Manager

A local, zero‑dependency web dashboard to find, overview and manage all your Claude Code sessions across every folder on your machine — including the ones you started and forgot.

Claude Code stores every session centrally under ~/.claude. Once you work across many projects, it gets hard to remember what you started, where, and whether it's still running. This tool reads that data (read‑only) and gives you one clean overview — as a list, a timeline, or a kanban board — with full‑text search and backup built in.

⚠️ Unofficial community project. Not affiliated with or endorsed by Anthropic.

---

More views

List + detail — all sessions grouped by project, with a rich detail panel:

Timeline — an absolute time axis showing when each session ran, with activity markers:

---

Why?

• You start a Claude Code session in some project, get interrupted, and forget it's still running.
• You can't remember which folder a session was in, or what it was about.
• You want a bird's‑eye view of everything you've worked on and its current state.

Claude Session Manager solves exactly that.

Features

• 🟢 Live status — see which sessions are actually running (🟢), idle (🟡) or closed (⚫), via a real process check.
• ☰ List + detail — sessions grouped by project; click for a rich detail panel:
  • 🎯 Goal (first prompt) and 📍 latest state (last prompt) — an instant summary, no AI call needed.
  • 💬 full prompt history, 🖼️ image gallery (screenshots pasted into the session), message counts, duration, branch, PID, folder.
• ⧗ Timeline (Gantt) — a single absolute time axis (years → months → days) showing when each session ran, with prompt markers for activity.
• ▦ Board (Kanban) — drag sessions across your own columns to organize your workflow.
• 🔍 Full‑text search across all transcripts — prompts and answers. Find the session by what was said in it, with highlighted snippets ranked by relevance.
• 💾 Backup (see below) — one‑click git push per project, plus automatic session‑history backup that survives Claude's auto‑cleanup.
• 🎴 Visual project filter (tiles) · 🏷️ status filter chips · ⌨️ keyboard navigation (/ search, ↑/↓ select, Esc clear).
• ▶ Resume a forgotten session in a new terminal, 📂 open its folder, 📝 add notes.
• 🌗 Light & dark theme, 🌍 multi‑language (English/German included), all configurable.
• 🔌 Config‑first — almost everything is driven by a settings schema (a stepping stone toward a plugin system).

How it works

It reads, read‑only, from your local ~/.claude directory:

Source	Used for
~/.claude/sessions/<pid>.json	running sessions (live status)
~/.claude/projects/*/*.jsonl	transcripts → titles, goal, prompts, images, timing, full‑text search
~/.claude/session-backup/ (own git repo)	backed‑up transcripts → also read to show archived (cleaned‑up) sessions

Your own data (kanban columns, notes) and settings live in separate sidecar files (~/.claude/session-manager-state.json, session-manager-config.json). The original Claude Code files are never modified.

Privacy

• 100% local. Nothing is uploaded anywhere — it's a plain Node HTTP server on localhost.
• The web UI is only served to your machine (default http://localhost:4317).
• Read‑only access to your sessions; writes go only to its own sidecar/config files.

Backup

The Backup tab has two independent parts:

1. Per‑project git status & push. For every folder your sessions live in, it shows whether it's safely backed up — backed up ✅, N unpushed ⬆, uncommitted changes ●, no remote ⚠, or not a git repo. One click pushes to the project's existing remote (uses your cached git credentials — no token needed), or commits & pushes all changes (with a confirmation).

2. Session‑history backup. Claude Code auto‑deletes local transcripts older than cleanupPeriodDays (default 30 days) — so old conversations, including the assistant's answers, silently disappear. This tool can copy the full transcripts into a git repo (incremental, on a schedule + on start). Backed‑up sessions stay readable inside the tool even after Claude deletes them (shown as archived). Set a private backupRemote in ⚙ to also push off‑machine.

Tip: also raise cleanupPeriodDays in ~/.claude/settings.json (e.g. 365) to keep transcripts locally longer — the backup repo is the real off‑disk safety net.

Requirements

• Node.js 18+ (tested on 22). No npm install, no dependencies.

Install & run

git clone https://github.com/xreader/ai-session-manager.git
cd ai-session-manager
node server.js

Then open http://localhost:4317.

Windows: double‑click start.bat (starts the server and opens your browser). Use stop.bat to stop it.

Custom port: PORT=5000 node server.js (or set it in the settings).

Configuration

Open ⚙ in the top‑right. Everything is stored in ~/.claude/session-manager-config.json.

• Language, theme (light/dark), start view, auto‑refresh interval, server port
• Kanban columns (freely definable)
• Terminal used for “Resume” — presets per platform with a live preview:
  • Windows: PowerShell, cmd, Windows Terminal
  • macOS: Terminal.app, iTerm
  • Linux: gnome‑terminal, konsole, xterm
  • …or a custom command with {cwd} and {id} placeholders.
• Backup: optional backupRemote (git URL for off‑machine push), auto‑backup interval (hours), and backup folder.

Adding a language

Copy lang/en.json, translate the values, set "_name" (e.g. "Français"), save as lang/fr.json — it appears in the language picker automatically. No code changes. PRs welcome!

Roadmap

• 🧠 Support more AI assistants, not just Claude Code — pluggable “session providers” for other coding agents/CLIs (e.g. tools that keep local session/transcript data).
• 🔔 Notifications for forgotten/idle running sessions.
• 🗜️ Compressed off‑site backup (gzip transcripts) so large histories push cleanly to GitHub.
• 📈 Activity pulse (GitHub‑style) instead of single dots in the timeline.
• 🧩 Plugin system — contribute views, columns, commands and settings via drop‑in modules.

Recently shipped: 🔍 full‑text search · 💾 per‑project git backup + session‑history backup · 🏷️ status filters · ⌨️ keyboard nav · 🌗 dark mode.

Ideas and feedback welcome — open an issue!

Contributing

Pull requests are welcome! 🎉

• Small, focused PRs are easiest to review.
• New translations are especially appreciated (see “Adding a language”).
• For larger changes, open an issue first to discuss the direction.

The whole app is intentionally tiny and dependency‑free: a single server.js, a single index.html, and lang/*.json. Easy to read, easy to hack.

License

MIT — do whatever you like, no warranty.