AI Import — Set Up a Workspace with AI

With AI Import you set up complete workspaces by having an AI (e.g. ChatGPT, Gemini, Claude) describe them in a simple text format — Pachurros Markup. You paste the result, check the preview, and import with one click.

Find it in the left sidebar under AI Import (route /w/<workspace>/ki-import).

In three steps

1. Copy the prompt — On the AI Import page, copy the system prompt (“Copy prompt” button) and give it to your AI.
2. Describe what you need — Tell the AI what to build, e.g. “Set up a workspace for a video project: a project, a tasks table with status and assignees, a Kanban view, a dashboard page, and a few to-dos.” The AI replies with Pachurros Markup.
3. Paste & import — Paste the markup into the input field → Preview (shows what will be created) → Import. You land directly on the first created item.

References (relations, project assignments, mentions, embedded views) are resolved automatically on import.

Two texts for your AI

The AI import page offers two blocks to copy — they do different things:

• System prompt — an instruction: “produce Pachurros markup, output only the markup”. The short route: copy it, hand it to your AI, say what you need.
• Full format reference — an explanation: the whole of Pachurros Markdown in one text. Prose, every block, the frame (tables, pages, views, metrics), linking, a complete example and a checklist. Meant for when your AI should actually understand the format: paste it once, then simply describe what you want. Also useful for models that get sloppy with terse instructions.

Both come with a copy button and a .md download. When the format changes, both change with it — copy them again when needed.

Export & round-trip

You can also export your existing workspace as Pachurros Markup — on the AI Import page via “Export workspace as markup”. This lets you hand existing content to an AI, have it extend or revise it, and import the result again.

The export covers: projects, tables (incl. relations, formulas, views with filter/sort/group, subtasks), pages (with mentions & embedded views), rich rows, sessions, boards and tasks. Computed columns (formula/lookup/rollup/count) are recalculated automatically on re-import rather than transferred as values.

Add or replace?

Import always creates content anew — re-importing the same export into the same workspace creates everything twice. To put the edited state in place of the old one instead, enable the “Empty workspace before import” option before importing: the workspace is emptied (with a warning and confirmation) and then repopulated — without duplicates. Alternatively, import into a fresh, empty workspace.

Basic principle

Every piece of content lives in a block that starts with ::: and ends with ::::

::: table Tasks
...
:::

Block types: table, page, project, session, board, view, tasks, row, template. An optional emoji may follow the name (::: table Tasks ✅).

The German names work just as well: tabelle/datenbank, seite/notiz, projekt, termin, tafel, ansicht, aufgaben, zeile/eintrag, vorlage. Indentation before the ::: and four instead of three colons are fine too.

When the AI is sloppy

AI output is rarely formally perfect. The importer is built so that no content disappears silently — worst case it lands as readable page text. Anything noticed is listed after the preview under “Beim Lesen aufgefallen”:

The AI’s mistake	What happens
Section without a closing :::	The rest is still read and assigned to that section — with a notice
Unknown section (::: calendar …)	Content lands as page text — with a notice
Table without - Name: type lines	Columns are derived from the header row, the type guessed from the column name (first column is always the title)
Table without outer |	Read the same as with them
Row with too few cells	Padded instead of shifted
::: inside a code block	Does not close the section
Answer wrapped in ```markdown	Unwrapped
Title heading repeated below the page title	Removed once instead of shown twice

Page text itself is standard Markdown (CommonMark + GFM) — the full list is under All Block Types.

Tables

Columns are declared with a type, then after --- a Markdown table holds the data. The header row must contain the column names.

::: table Tasks ✅
- Title: text
- Status: select (Open, In progress, Done)
- Due: date
- Project: relation → Projects
---
| Title    | Status      | Due        | Project         |
| Concept  | In progress | 2026-07-10 | Website Relaunch |
:::

• The first text column is the row’s title — relations and mentions point to it.
• Cell formats: date = YYYY-MM-DD, checkbox = true/false, relation/multiselect = multiple values separated by commas.

Column types

Type	Description / syntax
text, longtext	Short / long text
number	Number — optional (currency EUR) or (percent)
date, time	Date / time
checkbox	Yes/No
url, email, phone	Web address / email / phone
rating	Star rating
select (A, B, C)	Single choice with options
multiselect (A, B, C)	Multiple choice
relation → TargetTable	Link to another table
assignee	Assign a person — cell = name or email of a member
location	Place / address
formula (expression)	Computed field, e.g. formula (Budget - Spend)
lookup (via <relation> → <field>)	Pull a value from a linked table
rollup (via <relation> <agg> <field>)	Aggregate from a linked table (agg: sum, avg, min, max, count)

Examples of computed columns:

::: table Tasks
- Title: text
- Project: relation → Projects
- Owner: assignee
- Remaining: formula (Budget - Spend)
- Project budget: lookup (via Project → Budget)
- Total effort: rollup (via Project sum Budget)
---
| Title   | Project          | Owner         |
| Concept | Website Relaunch | anna@firm.com |
:::

lookup and rollup reference a relation column in the same table (via) pointing to a column of the target table.

Subtasks (nested rows)

A row becomes a subtask of the previous top-level row when its title starts with ↳ :

::: table Tasks
- Title: text
- Status: select (Open, Done)
---
| Title          | Status |
| Main task      | Open   |
| ↳ Step 1       | Open   |
| ↳ Step 2       | Done   |
:::

Views

A custom view of a table — with grouping, filter, sort and visible columns:

::: view By Status on Tasks
type: kanban
group: Status
sort: Due desc
filter: Status is Open
columns: Title, Status
:::

• type: table, kanban, calendar, gallery, list, timeline, chart
• group: column for Kanban grouping · calendar: date column for the calendar
• filter: <column> <op> <value> — op: is, is not, contains, not contains, empty, not empty, gt, lt, gte, lte · sort: <column> <asc | desc>
• columns: — visible columns (comma-separated)

Timeline (Gantt): type: timeline + start: <date column> + end: <date column>

Chart: type: chart + chart: bar | donut | line + group: <column> (X axis) + value: <number column> (optional) + agg: count | sum | avg | min | max

Embed a table or view on a page

A line of its own in page text embeds a table or a specific view:

@embed Tasks                          (whole table, default view)
@embed Tasks / By Status              (specific view)
@embed Tasks / Board | List | Calendar  (several views as tabs)

Separate multiple @embed lines with a blank line each.

Scope to the current row (@current)

In a view with filter: <Relation> is @current, the placeholder resolves at render time to the row whose record page the embed lives on. So one generic view automatically shows the linked entries on every row (e.g. “Tasks of this meeting”) — and a row created inside it gets the relation set automatically.

::: view Tasks of this Meeting on Tasks
type: table
filter: Meeting is @current
:::

On ordinary pages (no record context) such an embed is greyed out.

Pages

A rich-text page with headings, lists, tasks, quotes and code:

::: page Project note 📝
# Project note
Body text with **bold**, *italic*, `code` and [links](https://example.com).

- Bullet
1. Numbered list
- [ ] open task
- [x] done task
> Quote
---
:::

Special blocks

:::collapsible More details
Expandable section.
:::

:::columns2
Left column
---col---
Right column
:::

:::callout 💡
Highlighted note.
:::

:::toc
:::

:::equation
E = mc^2
:::

:::columns3 creates three columns (each separated by ---col---).

Nesting pages & assigning a project

At the very top of the ::: page block (before the content):

::: page Detailed concept
parent: Project note          (nests the page under "Project note")
project: Website Relaunch      (assigns it to a project)
# Detailed concept
...
:::

Mentions

Inside page text, link with [[…]]:

[[Tasks/Concept]]     → links a table row
[[Project note]]      → links a page

Rich rows (record pages)

Give a table row a full page that opens when you click the row — with body text, lists and even embedded views:

::: row Tasks / Concept
project: Website Relaunch   (optional — assigns the record page to a project)
# Concept details
Detailed description, checklists, embedded views (@embed) …
:::

::: row <Table> / <Row title> — the row is found by its title.

Row templates

A named template per table: new rows start with a predefined page structure and field values. Exactly one template can be default: true — it applies on the table’s “New row” button.

::: template Standard on Meetings
default: true
Status: To Do
## Decisions
-

## Action Items
@embed Tasks / Tasks of this Meeting

## Discussion
-
:::

• Leading key: value lines are either default: true or property defaults (e.g. Status: To Do pre-sets the “Status” column on creation).
• Placeholders: {{title}} (row title) and {{today}} (today’s date) are resolved on creation.
• Any @current embeds inside resolve to the new row automatically.

Dashboard controls

On a page with several embedded views (overview/dashboard):

@filter Projects                 (one button per row of the "Projects" table)
@filter Tasks / Status           (one button per option of the "Status" column)
@filter Tasks / Status (multi)   (allow multiple selection)
@button Tasks                    (button: create a new empty row + open it)
@button Tasks / Standard         (create a new row from the "Standard" template)

• @filter sets a page-wide filter. Every matching embed filters along automatically: in relation mode all embeds with a relation to the target table (e.g. “one button per project filters all views”), in select mode all embeds with a same-named select column.
• @button creates a new row on click (optionally from a row template) and opens it.
• Put each @filter/@button on its own line (separated by a blank line).

Metrics (dashboards)

One big number from a table — as its own block inside page text:

:::metric Open tasks
table: Tasks
agg: count
filter: Status is Open
form: none
:::

Key	Meaning
table:	the table the number comes from
agg:	count, sum, avg, min, max — short form agg: sum Amount
value:	number column (for everything except count)
filter:	<column> <is|is not|contains|not contains|empty|not empty|gt|lt|gte|lte> <value>, repeatable
form:	the picture above the number (see below). Without it, only the number is shown.
prefix: / suffix:	unit before or after
format:	plain, thousands, currency CHF, percent · decimals: auto|0|1|2|3
emphasis:	strong (lead number) or quiet (secondary number)

The picture above the number

• form: none — just the number. This is also the default when form: is missing.
• form: line · area · bars — a trend. Add axis: <date column>; without it the row’s creation date counts, and imported rows all land in the same month. Optional: range: 12m | 6m | 12w | 30d and cumulative: yes (each point shows everything up to then).
• form: progress · ring — share of a goal. Needs goal: <number> or goal: metric <name of another metric on the page>.
• form: delta — arrow with the gap: compare: previous (previous period, needs axis) or compare: goal.

Several tiles (grid)

Several metric: <title> groups in one block; grid: sets the tiles per row:

:::metric
grid: 3

metric: Total budget
table: Funding
agg: sum Amount
form: area
axis: Date
cumulative: yes

metric: Drawn down
table: Funding
agg: sum Amount
filter: Type is Drawdown
form: progress
goal: metric Total budget
emphasis: strong
:::

Calculated metric

Named building blocks and a formula on top — this is how row sets are combined:

:::metric Remaining budget
term Budget: table Funding, agg sum Amount, filter Type is Budget
term Drawn: table Funding, agg sum Amount, filter Type is Drawdown
formula: {Budget} - {Drawn}
:::

A building block may also point at another metric on the page: term Plan: metric Total budget.

Names must match

Tables and columns are connected by name on import. If a filter column can’t be found, the condition is dropped — the metric then counts more rows than intended.

Projects

A project groups tables, pages, sessions and boards:

::: project Website Relaunch 🚀
status: Active
color: #3b82f6
Rebuild of the company website (optional description, multi-line).
:::

Assignment: project: <ProjectName> as the first line of a table, page, session, board or row block.

The assignment decides where an item lives

In the sidebar every item has exactly one place — it never appears twice:

	Where it shows up
with project:	inside the project only, under Projects
without project:	under Workspace only, as a general item

So assign consistently: anything that belongs to a piece of work gets project:. Without it, the item ends up in the general area and looks unsorted there. Only genuinely cross-project things (contacts, templates, personal notes) are left unassigned on purpose.

Child items inherit the project from their parent — a subpage (parent:) does not need project:.

For the AI

Tell the AI explicitly to create a project and attach all content to it. The system prompt from the AI import (In three steps) already contains this rule — if you copy it, the AI assigns everything on its own.

Sessions

::: session Kickoff
type: Meeting
date: 2026-07-05
time: 14:00
location: Zurich
project: Website Relaunch
Description (optional).
:::

Boards

Each line becomes a sticky note on the board:

::: board Ideas
project: Website Relaunch
- First note
- Second note
:::

Tasks (to-do module)

Creates real entries in the Tasks module:

::: tasks
project: Website Relaunch
- [ ] Write proposal | due 2026-07-10 | @anna@firm.com
- [x] Send invoice
:::

- [ ] open, - [x] done. Optional per line: | due YYYY-MM-DD | @name/email.

No assignee, no task list

My Tasks only shows what is assigned to you. So give a task an owner via @name or @email wherever it is clear who takes it on — otherwise it shows up in nobody’s list.

The name or email must match a workspace member; unknown names are imported without an assignee. The same applies to to-do blocks on pages and to assignee columns in tables.

Full example

::: project Website Relaunch 🚀
status: Active
Rebuild of the company website.
:::

::: table Tasks ✅
project: Website Relaunch
- Title: text
- Status: select (Open, In progress, Done)
- Due: date
- Owner: assignee
---
| Title   | Status      | Due        | Owner         |
| Concept | In progress | 2026-07-10 | anna@firm.com |
:::

::: view By Status on Tasks
type: kanban
group: Status
:::

::: session Kickoff
type: Meeting
date: 2026-07-05
location: Zurich
project: Website Relaunch
:::

::: page Dashboard 📊
project: Website Relaunch
# Overview
Current tasks — see [[Tasks/Concept]].
@embed Tasks / By Status
:::

::: tasks
project: Website Relaunch
- [ ] Prepare kickoff | due 2026-07-04
:::

Limits

Some things can’t sensibly be created from text and are intentionally not included:

• Image, file and video uploads and media library content
• Board elements beyond sticky notes (shapes, frames, connection lines)
• Column description / required flag

Tips

• Use the preview: Before importing, the preview shows how many projects, tables, pages, sessions, boards and tasks will be created.
• Keep names consistent: Relations, project and view assignments work via exact names/titles — the AI should use them consistently.
• Work in stages: Build large setups across several imports; existing items can be referenced further by name.
• Keep the prompt current: The system prompt on the AI Import page always contains the full, up-to-date format reference — copy it again when needed.