Tasks everything you do.

A task is a snippet with a tag

There is no second app and no second database. A task is one of your snippets, marked as something to do. That sounds like an implementation detail and it is the whole design: everything that already works on a snippet works on a task.

Tag it, search it, cite it in a post, read it from any client, attach a file to it. The date is a tag. Done is a tag. The steps are the bullet points in the body. Nothing about a task is a field only one app understands.

And because a task is a snippet, your to-do list and your notes are searchable together. The task that says write the summary sits one tag away from every snippet you collected in order to write it.

Reminders, dates and steps

Give a task a date and it shows up when it is due. Break it into steps and tick them off as you go — the steps are ordinary bullets, so you can write them in the terminal, dictate them to an assistant, or type them in the app, and they are the same steps.

Closing a task does not delete it. It stays in the library as a record of what was done, next to the work it produced.

Assign it to an AI

Some tasks you do. Some you describe. Write what you want in plain language, tag it for an agent, and a session starts for that one task — it reads the request, looks through the snippets tagged for that project, does the work, and writes the result back into the task.

The result lands inside the task, under what you asked for, so the request and the answer stay in one place. Then the task closes itself. If it could not finish, it says what it needs and stays open for you.

1. Write the task. One or two sentences is usually enough.
2. Tag it for the agent — from your phone, mid-walk, with no terminal in hand.
3. It picks the task up, loads the context your tags point at, and works.
4. You come back to the result written into the task, and a closed item.

This is why the knowledge base matters. An agent handed a task with no context asks you everything; an agent pointed at the tags that hold the project already knows what was decided, what was tried, and why something was rejected.

The work product goes where the work goes

Whatever a task produces — your writing or an agent's — is a snippet in the same library. Tagged, searchable, citable in a post, readable by the next session that works on the same thing.

Nothing worth keeping is left in a chat window to be lost when the window closes. A finished task is a durable record: what was asked, what came back, and what was found along the way.

One interface, two halves

Tasks is a view of the same library, in the same app, with the same chrome — the same tag fields, the same cards, the same swipes, the same words. Moving between what you know and what you do is a tab, not a context switch.

It is also the same convention everywhere: because a task is just tags and Markdown, the web app, the terminal client and an AI client over MCP can all create one, date one, and close one. No client needs to be taught a new object.