One file.A working tool.

Try three ways to show the same numbers

A dashboard lets you choose. Slides tell it in order. A report ties text and numbers to one version. Below, all three use the same made-up dataset. See how the same numbers become three different reading experiences. What is available now?

Make your first file

The person who makes the file needs an AI tool that can read skills, and one preparation done once. The person who reads the file needs no licence, no platform, no server and nothing installed.

1. You have the numbers

   A spreadsheet, a CSV or a query result. That is the whole input. To try it with something finished, use the sample file in the repository: examples/01-hello-dashboard/salgsdata.csv.

2. You install the skills

   This is the only preparation, and it is done once. In RAIWork you install the Tún plugin from the marketplace. Without RAIWork, fetch the repository and run python3 setup.py, which puts the skills where your tool reads them. Nothing else has to be installed: the tools in the repository use only the Python standard library.

3. You ask for the file

   Say what the file must contain, not how to build it. For example: "Make a dashboard of salgsdata.csv with product as the dimension, amount as the measure, and region, product and channel as filters." The skill follows the recipe and produces one HTML file with the numbers inside. The code and the skills live on GitHub.

4. You check it

   Open the file in a browser and see that the numbers match the source. The repository has a check that does the same in a real browser and gives you a screenshot: tools/verify_dashboard.py. A clean build log is not evidence.

5. You share it

   Put the file in the folder or send the link. The reader needs no account and no licence. What the reader actually gets, and which previews show the file without running code, is under the limits.

One core, many kinds of work

In Old Norse, tún is the place where the farm gathers. Here it is a picture of what we want to build: a place for knowledge, reuse and people who make everyday reporting tasks easier. The question is the same every time: What does this file need to carry with it?

The core we want to build is Markdown skills (recipes your AI tool follows) plus reusable HTML, CSS and JavaScript. Skills describe the task and the checks. Templates give structure. Checks look at the data, accessibility and the result. Reuse what works and spend your time on the story you want to tell.

Choose how the work lives on

A basis for a decision can be frozen. A weekly overview should be renewed. A working tool may need fresh answers. The data agreement describes what travels with the file and how the next version arrives.

quarter/
├── dash.html
└── data.js

From Markdown skill to checked result

A skill is an editable recipe: source, template, checks and approval. Here is what one can look like:

A draft SKILL.md

Illustrative draft: a small recipe that can be read, changed and reused.

---
name: report-extract
description: Make a short report from a checked data extract.
---
1. State the sources, method, data time and data version.
2. Use a version-pinned HTML template. Freeze text, numbers and chart together.
3. Validate totals and conclusions. Try keyboard, mobile and delivery place.
4. Show the whole result to a person. Get approval before publishing.

Share an improved recipe and the next producer starts one step further along. Tún is meant to gather data agreements, reuse and checks around tools that already do the job well.

Fewer detours, more content

Think toolbox, not one big program. These are jobs we want to make easier: small deliveries that can skip needless Office and BI admin. Pick the story and let the format make it easier to use.

• The sales meeting → a story the reader can explore

  Let a slide deck carry interactive charts instead of screenshots. Switch the period, see the difference and go back to the numbers while the conversation goes on.

  When do you need more? (the sales meeting)

  PowerPoint's full editing environment and PPTX export are separate needs. Row-level security and live queries call for a suitable data solution.

• The decision memo → a report with claims you can check

  Let every important claim point to its exact table. The report's total of NOK 140m shows the pattern: text, numbers, chart and provenance follow the same version.

  When do you need more? (the decision memo)

  Simultaneous editing and change tracking may belong in a document tool. A new data version needs a new check of the report text.

• “What if?” → a small scenario calculator

  Let the reader change the quantity and unit price, and show the formula next to the answer: amount = quantity × unit price. With 120 units at NOK 250 each, the example comes to NOK 30,000. The assumptions are visible, not hidden in a cell.

  When do you need more? (“What if?”)

  Complex dependencies, Excel's calculation model and approved decision models need their own tools and tests.

• “What do I do now?” → a process map with work instructions

  Show the flow from intake to checking and delivery. Let each step open its responsibility, checklist and runbook in the same document. Overview first, details when they are needed.

  When do you need more? (“What do I do now?”)

  A readable process map is not the same as an engine that runs the process, simulates it or steers approvals.

• The project update → a room for the context

  Gather the timeline, milestones, decisions and attachments. Let the reader follow a decision from date to reasoning, instead of digging through meeting minutes and slides.

  When do you need more? (the project update)

  Joint editing, audit trails and different access levels may call for a connected project tool.

The limits that make the result trustworthy

A file cannot hold a secret. Everything that comes with it can be read by whoever gets the file. Service keys and other secrets never belong in generated HTML.

This is where the toolbox starts

Available now: the core is open source. It builds dashboards, cubes with millions of rows, process diagrams and slide decks as single files, and its skills tell your AI tool how. The three demos on this page are original HTML/CSS with made-up numbers.

Every example: they open in the browser, with no sign-in and nothing to install. jkorsvik.github.io/tun

Planned: skills, templates and checks from the community.

Licence: the core is free software under AGPL v3 or later. The documents you generate are yours, under any terms you like, by the Artefact Exception. It asks two things: keep the short source line that names Tún, which does not have to be visible, and do not make the document itself a rival generator. The code is on GitHub.

Want to follow along?

This is early. Do you have a small job that takes too much admin, an example to share, or an objection? Please get in touch, or open an issue on GitHub. The page has no sign-up form, tracking code or cookies. The email link opens your email program and sends nothing from this page.

Tell us you are interested

You can also copy the address: hei@tun.community