codenanny

Find which Claude Code session built that abandoned folder. Resume disconnected sessions with one paste. Local, MIT, no cloud.

Codenanny watches your local Claude Code transcripts and turns them into a navigable index of every session — every prompt, every file Claude wrote, edited, or read, and every file-creating Bash command (cat>, tee, mkdir, etc.) — so you can:

• Resume a disconnected session with one click. Paste-ready markdown bundle with the last N turns + files created/edited + files read for context.
• Find which session built a folder. Paste a disk path, see the sessions that touched files in it, ranked by recency, with action chips (✏️ edit ×10, 👁️ read ×6, $→ bash-write ×2).
• See what a session actually did. A unified timeline interleaves prompts and file ops in conversation order. No more "what was I doing here?"

Runs entirely on your machine. No inference, no cloud, no account. The transcript data already lives in ~/.claude/projects/; codenanny just makes it queryable.

Install

npm install -g codenanny

Requires Node 20+.

Quick start

codenanny serve
# → indexing ~/.claude/projects/ ...
# → live at http://localhost:7700

Open the URL. Four views in the sidebar:

• Sessions — every session, newest first. Click one for the timeline.
• Media — every file op across every session.
• Projects — group sessions by working directory.
• Find — paste a disk path (e.g. /home/me/projects/old-thing/) to see which sessions touched files in it.

The 📋 Resume button on the session header copies a paste-ready bundle to your clipboard. Drop it into a new Claude session and continue from where you left off.

CLI

codenanny serve                        # live web UI (default port 7700)
codenanny ingest                       # one-shot index, no server
codenanny export                       # static HTML+JSON bundle
codenanny resume <session-id> --copy   # resume bundle to clipboard via OSC52
codenanny wizard                       # browser-based setup
codenanny help                         # full options

Common flags:

codenanny serve --port 7700 --db ./codenanny.db --src ~/.claude/projects --watch
codenanny resume <id> --turns 10 --json

How it works

• Walks ~/.claude/projects/*.jsonl (Claude Code's transcript directory).
• For each session: extracts user + assistant turns, every tool_use block (Write, Edit, MultiEdit, NotebookEdit, Read), and heuristically extracts file paths from Bash commands (>, >>, tee, mkdir, touch, cp, mv).
• Stores in a local sqlite + FTS5 database. Watches for new transcripts via chokidar with a 750ms debounce.
• Exposes a JSON HTTP API + a vanilla-JS UI (no framework). The UI subscribes to a server-sent-events stream for live updates.

What's coming

• Multi-agent adapters: Gemini CLI, Codex CLI, Aider.
• Topic-based session clustering (group sessions by intent, not just by working directory).
• Daemon mode polish: start/stop/status, browser-open, systemd/launchd templates.
• Opt-in multi-user mode (users table already exists).

Links

• Source + issues: https://github.com/nobleglitch/codenanny
• Core library (for plugkit hosts): @codenanny/core

License

MIT. Created by nobleglitch.