Memo

Memo Viewer

Reads the local Claude Code transcripts in ~/.claude/projects/ — every session's Bash commands, reasons, thinking, pass/fail status, and cost, plus the skills/subagents each session used.

Commands at a glance

Pick the one matching intent; details for each below. All three accept the same scoping flags (--here, --project, --days, --failed-only, --limit, --source). Run from the user's current directory; default to --here to scope to their project.

How to run

Run from the user's current directory (do not cd elsewhere). By default, scope to the project the user is currently in with --here — it keeps only sessions whose working directory is the current project (or a subdirectory):

memo --open --here

If memo isn't on PATH, the module form works from any directory once the package is pip-installed:

python3 -m memo --open --here

--here keys on the current directory — so it must run from the user's project root (where their Claude Code session is working), not from memo's own repo. That's why there's no cd above.

Narrowing

--here is the default scoping. Combine or replace it as the user asks:

• All projects (not just the current one) → drop --here: memo --open
• A different/named project → --here <DIR> or --project <substring-of-cwd> (e.g. --project quizlab)
• Only sessions with failures → --failed-only
• Recent activity only → --days <N> (e.g. --days 7)
• Most recent N only → --limit <N>
• A specific AI tool → --source claude|copilot|codex|gemini|cursor (repeatable; default: all)
• Custom output path → --out <path> (default dist/index.html)

Example — "failed commands from the last week in this project":

memo --open --here --failed-only --days 7

After running

Report the output path and the session/command counts the tool prints. The page opens in the browser with live search, source/project filters, a "failed only" toggle, expandable sessions, a 💰 Cost page, and per-project skills/subagents. The tool does all the parsing — do not parse the JSONL yourself.

Note: Claude's thinking text is only sometimes persisted in transcripts; when it is cached/absent the viewer omits it (the per-command description is the reliable "why").

List — see the sessions for this project

When the user wants to see which past sessions exist (rather than open the HTML), use the list subcommand. It prints one newest-first row per session — short id, date, command/fail counts, and title — straight to the terminal. Return that text; don't open a browser.

# Sessions for the current project:
memo list --here

# Across all projects, last 7 days, failures only:
memo list --failed-only --days 7

Each row leads with a short session-id prefix that memo digest <prefix> accepts, so list then digest is the natural pair: list to find the session, digest to pull it into the current conversation.

Digest — load a past session into this one

When the user wants to recall or carry over what happened in an earlier session (not browse the HTML), use the digest subcommand. It prints a compact markdown primer — goal, key decisions, the commands that ran (and which failed), and files touched — small enough to drop straight into the current conversation as context. The subcommand's output is the primer; return it to the user / read it into context rather than opening a browser.

# Most recent session in the current project:
memo digest --here --last

# A specific session by a substring of its title (newest match wins).
# Put the selector FIRST, before --here, so it isn't read as --here's directory:
memo digest "auth refactor" --here

# A specific session by id, across all projects:
memo digest 0c1f2e3a-....

All generate filters apply (--here, --project, --days, --source, …) to narrow which sessions are candidates. Add --out <path> to write the markdown to a file instead of stdout.

Gotcha: --here takes an optional directory, so memo digest --here "auth" reads "auth" as the directory. Always pass the title selector before --here (or use --project <substr> instead, which has no such ambiguity).

This complements — it does not replace — Claude Code's native claude --resume <id>: resume re-enters the whole old conversation (and its full context); a digest is a small one-time snapshot you pull into a different session.