pi-session-list0.2.1

pi-session-list

A small pi package for browsing your old sessions without leaving the one you're in. One TypeScript file, no dependencies.

pi saves every conversation as a session file, but the built-in ways to look at them are the --resume picker at startup and /resume in a running session. This adds a search-first picker and two model-callable tools:

Install

pi install npm:pi-session-list

From git, or for one invocation only:

pi install git:github.com/armagg/pi-sessions
pi -e npm:pi-session-list

Pull updates with pi update --extensions.

/sessions

/sessions                     # 20 most recent sessions for this project
/sessions all                 # every project
/sessions today               # only sessions touched today
/sessions all 50 insomnix     # every project, up to 50, matching "insomnix"
/sessions 7d ramzinex delta   # last 7 days, matching "ramzinex delta"

Tokens are positional-free: all sets the scope, a bare number sets the limit, today / yesterday / week / 30m / 24h / 7d / 2w / 2026-10-01 sets a lower time bound, and anything else becomes the search text. A row matches when the text appears in its id, name, project path, or first message — and otherwise when it appears anywhere in the transcript, in which case the row shows the matched passage instead of the first message.

Each row shows relative time, id prefix, session name, message count, project (when the scope is wider than one project), and the first message. Picking a row offers:

• Resume this session — switches the running pi to it (switchSession), across projects too. Interactive mode only.
• Preview last messages — the last 12 messages in an editor pane, closed without side effects.
• Copy file path / Copy pi --session … command — to the clipboard, so you can open the same session in another terminal.

A query that matches exactly one session skips the list and opens its actions directly, so /sessions 01a15886 is a two-keystroke resume.

/resume is still the better tool for renaming or deleting sessions; this one is for finding a session by what it was about and jumping straight into it.

Tools

recent_sessions — listing with filters:

session_transcript — read one session:

Output is the session's active branch, so compaction summaries appear where pi put them and abandoned branches stay out.

Session 01a15886-b1c2-77bd-a78e-017d52b6951b
Project: /Users/omid/my-project/pi-sessions
File: /Users/omid/.pi/agent/sessions/…jsonl
Messages: 99 on the active branch (104 message entries in the file, including other branches), showing 2 at indices 95-98
----------------------------------------
[96] assistant · 2026-10-20 09:18 (1 thinking)
  ↳ tool bash {"command":"tsc --noEmit"}
  ↳ result bash ok: tsc done

showing N counts rendered messages, so it always matches the body. A tool result is grouped with the message that produced it: a page never starts with an orphaned ↳ result, and limit/offset count messages rather than continuation lines. When recent_sessions and session_transcript disagree about a count, that is why — the listing counts message entries in the file (all branches), the transcript counts the active branch.

Notes

• Sessions are read through the host's SessionManager, so --session-dir, sessionDir, named sessions, and the usual cwd rules all apply. Listing every project reads every session file: ~0.5s for 170 sessions / 300MB here.
• The tool is called recent_sessions, not session_list, so it does not collide with pi-total-recall. If you run both, session_transcript and session_read are similarly independent.
• No runtime dependencies. @earendil-works/pi-coding-agent and typebox are host-provided peers.

Development

npm install
npm run typecheck
pi -e ./extensions/sessions.ts

Releases are published to npm by .github/workflows/publish.yml on a v* tag, which is also what makes the package show up in the Pi package gallery (the pi-package keyword is what the gallery indexes).

License

MIT