The daily note: you write once, in the shape you already use.

One page per day

There are no documents to create and no folders to pick: there is a day, and today is already open. Days scroll straight back, and each one saves itself while you type.

• Focus mode to edit a single day without the noise of the others, and hide weekends if you do not work Saturdays.
• Go to date to jump to a specific day, and the arrows to reach the newest or the oldest one.
• Search the current view with ⌘F, or every day — by text or by label — from the same box.
• On the right, the intelligence panel for the day you are standing on: its labels, its tasks, its decisions and its milestones.

And there are two ways to write into today's note without opening the tab:

Bullets, outlines and indentation: the shape is the data

The editor is block-based, like the one in documents: bullets, numbered lists, checklists, headings and plain lines. For the analysis, the marker does not matter: what matters is what each line hangs from.

• A line with no bullet opens a topic. It is the heading: the client, the person or the project.
• A bullet is an item of that topic. With `-`, `*` or `•`, with or without a space after the dash.
• Tab pushes it one level in; Shift+Tab pulls it out. Indentation is what builds the tree.
• If you type fast and would rather not indent, repeating the dash does the same: `-` is one level, `--` is two, `---` is three.
• Numbered lists and checklists count exactly like bullets. Use whichever reads better to you.

A well-formed note looks like this:

Acme
- meeting with Ana about the renewal
-- we have to send her the proposal on Friday
-- legal reviews the contract
- we decided to bill per milestone, not per hour

Internal
- finished the database migration
- pending: pick the release date

Inheritance: every line is read with its branch

When a line goes out to be analysed it does not go alone: it goes with the full chain of its parents as context. The deepest bullet in the example above arrives as "we have to send her the proposal on Friday", with "Acme" and "Ana" hanging off it.

Three behaviours follow from that, and they are worth knowing:

1. 1
   The parent's entities flow downThe person, organization or project sitting in the heading can be attached to the bullet's task, spelled exactly as it is written above. That is why the name goes in once, spelled right, instead of five times spelled wrong.
2. 2
   A parent with real children becomes context"Acme open items" or "Ana, finish this week" do not become tasks: the real item lives in the bullets underneath, and the line above only groups them. Without this rule, every heading would leave you a phantom task with no object.
3. 3
   Unless the children are parts, not items"Review the whole contract" with "section 3" and "section 4" underneath is still a task: the children are pieces of the same object, not independent assignments.

The criterion behind all of it: a task has to make sense without reading the bullets that follow it. If you need to look further down to know what to do, that line is context.

Predictive typing: your knowledge, as you write

From the third character on, what you are typing is matched against the names in your Atlas — people, projects, organizations and labels — and a small list appears next to the cursor.

• It matches the canonical name and the aliases too: if you saved "Mari" as an alias of Mariana Falcón, typing "Mari" offers you Mariana.
• It understands compound names: "carlos cast" already offers "Carlos Castro", because it looks at up to the three words before the cursor.
• It also catches a word in the middle: "falcon" finds "Mariana Falcón".
• ↓ ↑ to move, Enter or Tab to insert, Esc to dismiss. If you already typed the full name it does not appear: there is nothing to correct.
• It never shows up on closed days — that day is already processed.

The same box shows up in documents, where it also completes formula variables when you type `{{`. See Memory and documents.

What gets flagged as you write · Hobby

Detection runs on top of what you write without touching it: it lays flags over your text, it never rewrites it. Only new or edited lines are re-analysed, so the note is not reprocessed on every keystroke.

Every line with content falls into one of four classes, and the first three leave a numbered flag over the text:

In parallel, the panel on the right builds the labels of the day: people, projects, organizations and free labels, each linked to its Atlas entry. When a name it does not recognize shows up, it asks — who is "Ana"? — and you resolve it by linking to someone who already exists or creating them right there.

Everything is reversible before you close:

• The crossed-out eye dismisses a task, a decision, a milestone or a label. It stops counting for the close, and your text stays intact: what was surplus was the flag, not the line.
• Whatever you dismissed collects under Dismissed, and comes back with one click.
• What is already confirmed in the Atlas beats what the model guessed: if "Acme" is stored as an organization, it will not be treated as a person because one sentence made it look like a name.

How it reads time

Dates are read in plain words and resolved against the day of the note, not against today. That is what keeps reopening last Friday's note on a Tuesday from shifting every date: "Monday" in that note still means the Monday after that Friday.

A date sitting in a line flagged as a task becomes its due date. After that there are two ways to change it without rewriting the line: drag the provisional task onto the calendar — that date overrides the one in the text — or set it while closing the day, with the +1, +3 and +7 day shortcuts.

Two details that prevent surprises: time words ("Monday", "week", "yesterday") are never taken as entity names, and a task's urgency — overdue, today, this week — is worked out in your local time.

Reminders: a line that also pings you

A date makes a task due on a day. An hour does something else: it makes something reach you at that moment, whether the app is open or not.

There are two ways to write one, and both count the same:

• With the `/` → Reminder menu, which writes the shape already built, rounded to the next half hour: `⏰ 15:30 — llamar al contador`. Change the time and the text and you are done.
• By hand, in Spanish: "recordame mañana a las 9 llamar al contador". It understands clock times and also en un minuto, en dos horas or dentro de 3 días.

With no time there is no reminder: that is a task with a date, and everything above already covers it. The hour is exactly what creates something to announce.

They can also be asked for over chat, without opening the note: see WhatsApp and Telegram.

From a flag to a labelled task

While the day is open, every line flagged T lives as a provisional task: it shows up in the Tasks tab, greyed out and read-only, with a countdown of the days it has left. It already carries the labels it inherited from its branch and the date read from the text.

Provisional tasks are saved to your account, so they survive a reload and look the same from your phone. They last seven days: if you have not closed the day by then, they drop on their own.

Closing the day is what makes them real, and it is a review screen, not a blind button:

• Three columns — Tasks, Definitions and Milestones — with everything detected, and you can drag an item from one to another if it landed wrong.
• It groups by person, project, organization or calendar, so you can review client by client instead of down a flat list.
• Each item gets a date in one click, or stays undated. A task that is too big can be split into several with AI.
• Every item needs at least one label — person, project or organization — to be saved: a task with no owner and no project is useless to anyone a month from now.

On confirm, tasks are created with their labels and their date, definitions and milestones are recorded, new entities are registered in the Atlas and the day locks. You can reopen it whenever you want, and it can also close itself: past days left open close when midnight passes — validated when you open the app, not as a background job — if you enable it in the workspace settings.

What happens to those tasks afterwards — how they group into deliverables, how they look by person or by project — is in Tasks and detection.

Seven rules for writing the note

• One item per line. Two commitments in one sentence come out as a single task.
• The proper name on top, once. Everything hanging off that line inherits it.
• Make the task stand alone. "Review the Acme contract" is a task; "review that" is a note to yourself for the next two hours.
• Say the date in words. "on Friday", "next week", "November 15": they all resolve against the day of the note.
• Accept the suggestion with Enter. It is the difference between adding to an entry and creating a near-copy of it.
• Dismiss with the eye, not by deleting the line. What is surplus is the flag; the text is your record of the day.
• Close the day. It is the only gesture that turns any of this into tasks, entries and reports. Until you do, nothing exists.