Skip to main content

Quick start

Your first ten minutes. By the end you'll have edited a markdown note, built a database with a view, embedded a plugin, and talked to an agent.

1. Open the app​

Launch the app. It connects to the local agent_core backend on http://127.0.0.1:9090 and opens your workspace. Everything you do is stored locally — notes as markdown files, structured data in local SQLite.

2. Create and edit a markdown note​

Open the editor and edit a note. The editor renders markdown in Live Preview: as you type, special blocks turn into live widgets.

The note is plain markdown and stays the source of truth. Try a checkbox task — the editor understands the Obsidian-style task vocabulary:

- [ ] Draft the spec 📅 2026-07-01 ⏫ #proto
- [x] Sketch the idea ✅ 2026-06-01

📅 is a due date, ⏫ is a priority, #proto a tag, and ticking a box stamps a ✅ done-date automatically.

3. Make a database and a view​

Databases are Notion-style tables on SQLite. Create one and give it a couple of typed fields — for example a text title, a select status with choices todo / doing / done, and a date due field.

Add a few rows. Then look at the same data through different views — the same rows, different lenses. The bundled view plugins each show one axis of the data:

  • kanban — group by a status field
  • smart_calendar — a time axis (events)
  • smart_chart — a number over time

4. Embed a plugin in a note​

Inside a markdown note you can embed a view with a view fence. The fence names the data source, the plugin, a height, and optional key=value params:

```view db:tasks kanban 360 group=status
```
```view db:prices smart_chart 360 y=price x=at filter=symbol:TSLA tf=hour
```

The fence forwards source= plus the params to the plugin as URL params. The embedded view reads and writes your database live — edits made in the plugin show up in the grid (and vice-versa), because every surface talks to the same data contract. The same plugins also run standalone in the plugins dock with their own source pickers.

5. Talk to an agent in the console​

Open the agent console and send a message. The agent is LLM-driven and can read your notes and databases through its tools, then answer or take an action.

Two lightweight ways to put an agent right inside a note:

  • An ask block — a fenced block whose info-string is ask [agent]. Press the ▶ at the end of the fence and the channel agent answers, writing its reply as a > blockquote right after the fence:

    ```ask assistant
    Which tasks in this project aren't done yet?
    ```

  • A task binding — attach 🤖 <agent> or ⚙️ <cap>{params} to a task line; press ▶ to run it. The answer is written under the task and the task is marked done.

- [ ] Summarise this channel 🤖 assistant
- [ ] TSLA quote ⚙️ tool.stock_chart{symbol: TSLA, range: 1mo}

Nothing runs on open or on sync — an explicit ▶ click (or a POST /md/run) is the gate, so a shared note can carry these directives safely.

Next steps​