Agent Logbook

Browse, read, resume and measure the Claude Code and Codex sessions already on your machine. Everything is read from local transcript files. Nothing is uploaded, and no account is needed.

Website · Changelog · Privacy

What it does

Read your chats. Every session, grouped by project, with the full conversation rendered the way you saw it: markdown, images, tool calls, and the file diffs each edit made. Search across every transcript, not just titles.

Pick one back up. Resume opens the chat in a terminal, the Claude Code extension or Claude Desktop, and reveals the terminal already running it rather than starting a second one. Branch forks a session with its full context; Branch with a brief starts a fresh one from a written handover instead.

See the sub-agents. Background agents keep their own transcripts. Each one appears as a card where it was launched, with the brief it was given and what it reported, and opens as its own readable thread. Nested agents are shown as a tree, since an agent can spawn its own.

Know what it cost. A dashboard over your whole history - tokens, spend, cache efficiency, models, projects, files, working hours - plus per-chat stats including context growth and prompt-cache expiry. A shareable PNG card summarises a day, a week or all of it.

Codex too. Codex sessions are listed, read, searched and resumed alongside Claude's. Rollouts that are imported copies of Claude transcripts are skipped rather than counted twice.

Requirements

• VS Code 1.94 or later, or any editor built on it - Cursor, Windsurf, VSCodium and the rest.
• claude on your PATH to resume Claude sessions, codex for Codex ones.
• Plan-usage bars use the credentials Claude Code already stores: the macOS Keychain, or ~/.claude/.credentials.json on Linux and Windows.

Copying the share card to the clipboard needs wl-copy or xclip on Linux; Save PNG works everywhere.

What it reads and writes

It is worth being explicit, because two of these would look alarming if you found them rather than read them here.

Reads ~/.claude/projects and ~/.codex/sessions for transcripts, ~/.claude/**/memory and ~/.codex/memories for notes, and your Claude OAuth token from where Claude Code keeps it (the macOS Keychain, or its credentials file elsewhere) so it can ask Anthropic for your current plan usage - the same request Claude Code itself makes. The token is never stored or sent anywhere else.

Writes its own cache and flags under ~/.claude, and appends a custom-title line to a Claude transcript when you rename a chat, which is how Claude Code reads the new name back. Renaming a Codex chat writes nothing to the transcript. Deleting a chat moves the file to ~/.claude/trashed-sessions rather than unlinking it.

Settings

The Settings tab covers the quota bars across the top of the panel - which windows they show, in what order, in what colour - along with list ordering, where Resume opens, and what an export contains. Every control writes the same setting the editor's own settings page does.

Licence

MIT.