Plan in plain files. Get it done with your agents.

litpm is a local tracker for your own work: every task is a markdown file in a folder you own. Mention an agent and it plans, drafts and checks with you, and litpm stays out of its way. Nothing hidden, nothing to host.

A personal task list in litpm: a talk, a kitchen renovation and a book club as projects, with errands and due dates

Your week, in files you own.

A tracker that is a folder. The task on the left is the file on the right, and either one can change first. Grep it, sync it, back it up like anything else you own.

A task page in litpm: the talk outline, its done-when list, properties, a comment from Charlie and Claude Code's reply

---
title: Write the talk outline
status: done
priority: high
assignee: Charlie
due: '2026-09-20'
---
Three parts, one story. Start from the ticket that
changed how the team works, end on what stayed the
same.

## Done when

- [x] Three parts named
- [x] One demo picked
- [x] Under 30 minutes on paper

## Comments

### 2026-09-16 00:29 - Charlie

@claude draft the outline as three parts with one
line each, and suggest which ticket to use for the
demo.

### 2026-09-16 00:33 - claude

Draft outline, three parts, one line each. ME-1 is
marked done and the done-when boxes are ticked.

Edit anywhere, see it live

Open a task in the app or in any editor. The file is the truth, and whichever side you change, the other follows.

Projects are folders

A talk, a renovation, a book club: each is a folder with its tasks inside, milestones and progress read from what is there.

Readable without litpm

Markdown with a little frontmatter. Nothing to export, nothing to lose when you change tools.

An agent, when you want one.

Mention an agent in a comment the way you would a friend who is good at this. It reads the task, drafts, plans or checks, and answers in the same thread under its own name. Open the run to watch it think and to approve what it asks.

Comments on a task: Charlie mentions @Claude Code asking for an outline, and Claude Code answers in the thread

The run behind that reply: the prompt, the tools folded away, and the outline

Any agent, by name

Claude Code, Codex, Grok, Cursor, Gemini and more. Pick the one you already pay for.

Choose how much it may do

Supervised asks before every change. Full access never asks. Approvals and questions land in the thread, and litpm shows what is waiting for you.

Minimal by design

litpm doesn't wrap the agent in a framework or steer how it works. It starts the program you already run, with your own login, and hands it one skill: the litpm CLI.

One executable, or the app

bunx litpm in a terminal, or the desktop app for Mac, Windows and Linux. Nothing else to install.

A folder is enough

No account, no server, no sync service. Point litpm at a folder and it is a workspace.

A CLI your agents already speak

Every run gets the litpm command in its skill, so an agent lists, reads, comments on and closes tasks the same way you do in the app.