AI at Klaro Cards

Klaro Cards is built by a small team that uses AI every day: to write, to translate, to program. We do not use it to run your projects, and we hand nothing you put in your cards to a model.

This page says where that line falls, part by part. It is also where the DALIA badge on our articles sends you.

In short

• Every article on the blog declares how much AI went into writing it, on the DALIA scale, and wears it as a badge.
• The guides and the pages of this site are written by people, with AI assistance, and read by a person before they go live.
• The product carries no AI feature today. Nothing reads your cards in the background.
• Support is a person answering you.
• We never send your project data to a model, and what we do send is not used to train one.

The DALIA scale

DALIA is a six-level scale for declaring how much AI went into a piece of work. It is a self-declaration: the scale provides the words, it audits nobody. The specification is published under a Creative Commons BY-SA 4.0 licence.

• D0, manual: 100% human. AI is not used at all.
• D1, tool-assisted: AI analyses in the background. The human acts and decides.
• D2, AI-augmented: AI writes small, specific elements without changing the meaning.
• D3, guided creation: human and AI teamwork, each reworking what the other brings.
• D4, AI-generated, human-directed: the AI does the work, the human gives the directions and chooses the result.
• D5, fully autonomous: 100% AI, and nobody reads the result before it is published.

Nothing we publish is D5. Everything we publish is read.

Where the badge is, and what its absence means

The badge sits on an article under the title, and on the page of a series. Hovering it gives the definition of the level and a link back here.

An article without a badge was published before we began declaring, and we are working back through the archive. No badge does not mean D0: read it as "not declared yet".

An article and its twin in the other language each carry their own level. A translation made by a machine and then reviewed is a different piece of work from the original, and it says so.

The blog, and the pages of this site

The subject, the argument and the decision to publish are ours. Research is checked at the source, and a figure or a legal claim carries its link. Drafting is where the share of AI varies from one article to the next, which is exactly what the badge declares. Every article is read line by line before it goes out.

The pages of the site itself (what the product does, the prices, the comparisons) carry no badge: they are not dated articles, and they change continuously. They are written the same way as the blog.

The guides

guides.klaro.cards documents the product. A guide is written from the source code and from how the feature actually behaves, with AI assistance, then read by somebody who knows that feature. The translations take the same route. When a guide and the product disagree, the product is right and the guide is at fault: tell us.

The software

Part of the code of Klaro Cards is written with an AI coding assistant. No line reaches you unread: every change is reviewed by a developer, has to pass the test suites, and ships through the same release procedure as the rest.

Inside the product

There is no AI feature in Klaro Cards today. No automatic sorting, no generated text in your cards, no model watching your boards. If that changes, this page will say so, and the date at the foot of it will move.

The MCP server is the one exception, and you are the one who opens it. Klaro Cards exposes an MCP server, which lets you connect an assistant of your own (Claude, ChatGPT, a code editor, your own script) to your projects. When you do:

• the assistant acts in your name, with your account's rights, on the projects you give it access to;
• the model is the one your client chose, on the terms of whoever runs it rather than ours;
• we send nothing on our own initiative: a card is read because your assistant asked for it.

Nothing is on by default. Until you create a connection, there is none.

Your data, and model training

We do not send the content of your projects to a model provider. What we send is ours: our texts while we write them, our code while we work on it, our guides. We use assistants of the Claude family (Anthropic) for that, on a professional plan whose contract excludes training models on what we send.

Support is human end to end. When you write to us, a person reads you and a person answers.

Questions

Write to info@klaro.cards. If a badge on one of our articles looks wrong to you, that is worth a message too: a self-declared scale only holds if the people reading it can push back.