Tell your AI agent what you have to do. It captures, clarifies, reminds and helps you choose — following Getting Things Done and the rules you teach it.
A howware: your know-how, put to work by AI. The know-how here is Getting Things Done plus your own rules — what is urgent, which areas your life has, when you review. There is no app: you talk, the agent follows the protocol, a small tool keeps the dates exact, and a memory you own keeps everything.
› Note: call the accountant, pay the electricity and buy a gift for mom.
✓ Noted (3). Inbox: 3.
› Let's clarify.
Pay electricity → next action "Pay in the bank app" · Home · computer · 10 min · due Sat Oct 10, hard ·
repeats monthly on the 10th
Call the accountant → waiting? or next action "Call to ask for the tax return" · Work · phone
Gift for mom → project "Mom's birthday" · first step "Ask my sister what she'd like"
Shall I apply it?
› What do I do now? I have 20 minutes, I'm at the computer, low energy.
1. Pay electricity — overdue 2 days, hard deadline, 10 min.
2. …
| You say | It does |
|---|---|
| "Note…" | Captures to the inbox, no questions asked |
| "Let's clarify" | Triage GTD-style: next action (a verb), project, waiting on whom, someday, area, context, estimate, energy, hard or soft deadline, recurrence — proposed in one message, applied when you approve |
| "What do I do now?" | Asks your time, place and energy, applies your rules and suggests three options with reasons |
| "My work to-dos" | Lists by area, context, project or due |
| "Done" · "Move it to Friday" · "Forget it" | Completes (recurring tasks schedule the next one), postpones (counting how many times) or drops |
| (every session) | Reminds you of what is overdue until it's resolved; after 3 postponements asks whether to shrink, delegate or drop it |
| (your review day) | Walks you through the GTD weekly review: inbox to zero, overdue, projects without a next action, waiting, stale, someday |
| "Bank things are always urgent" | Stores a rule in your words and applies it — and proposes new rules from your history, which you approve |
Areas are yours. It starts with none: when something needs an area, the agent proposes one and you approve it. Contexts start as computer, phone, errands and home; you can add more.
| Part | Here |
|---|---|
| Protocol | skill/todo/SKILL.md — GTD and the rules for the agent, in plain words |
| Tool | todo.py — only what must be exact: today's date and weekday, due dates, recurrences (the 10th, the last Friday, every Tuesday…), overdue counts, states and history |
| Memory | The Long-Term Memory for AI adapter — your tasks and your rules, in a local file, with history |
The agent never guesses dates: "tomorrow at 10" is turned into a date with the tool's now and confirmed back
("Tuesday October 13 at 10:00").
git clone https://github.com/howware/long-term-memory-for-AI
python long-term-memory-for-AI/tools/install_skill.py --target ~/.claude/skills/project-memory
git clone https://github.com/howware/todo
python todo/tools/install_skill.pyCreate a folder for your to-dos, open Claude Code there (add the persistent-memory instruction from
Long-Term Memory for AI to its
CLAUDE.md) and say "I want to keep my to-dos here."
With your journal. todo can share its memory with bitacora (habits,
notes, collections): install both in the same folder, and the things to do you mention while writing your day land
in your inbox. Neither needs the other.
| Designed to stay on your computer | What the AI agent sees |
|---|---|
| The memory file with your tasks, history and rules | Task titles, next actions, areas, dues and history when it queries them; your rules; what you type |
What it cannot do or guarantee:
- No push notifications. Reminders appear when you open a conversation. (A calendar export for hard deadlines is planned.)
- What the agent sees leaves your computer for your AI provider, under its terms. Don't put passwords or secrets in a task.
- The memory file is not encrypted.
- Not yet tested in a real conversation on any host;
howware.jsonlists hosts only once they are tested.
git clone https://github.com/howware/long-term-memory-for-AI ../long-term-memory-for-AI
python -m unittest discover -s tests -v
git ls-files -z | xargs -0 python tools/check_private_data.pyTests use a temporary memory and a fixed clock (--now). See AGENTS.md and how this howware was
built with howware-builder: docs/building.md.
Apache 2.0. Not affiliated with David Allen or the David Allen Company; Getting Things Done® is their trademark.