There were a few use cases shown, and this was one of them: The Impact No. 01, a mechanical document editor.

I clicked on it and, wow. I do not know what it does under the hood, to be very honest. I have not explored it deeply yet. But as an object, as an interface, as a thing sitting inside the browser, it is just beautiful.

You can assemble the machine. You can switch into a cutaway view and see the mechanism. You can inspect it, turn on the sound, and type on your computer while hearing the keys hit. The mechanical parts move, the page responds, and suddenly writing feels physical again.

That is what struck me. We spend so much time inside flat, silent, disposable interfaces that when something like this appears, it reminds you that software can still have craft. It can have mood. It can make the machine feel present again.

I still need to spend proper time with Kimi Swarm and understand what it is trying to become. But this particular use case is worth seeing, even if only for the sheer beauty of the experience.

Useful links:

• Kimi
• The Impact No. 01 mechanical document editor

It is called book-to-skill, and it turns technical books, PDFs, document folders, and collections of sources into reusable agent skills.

A skill is basically a set of instructions and reference material that an AI agent can load when needed. When you invoke the skill, you are telling the assistant: this is the subject matter, this is the domain, and this is the expert frame you should use.

You could have a skill for copywriting. A skill for legal analysis. A skill for drafting rental agreements. A skill for accounting. A skill for chemistry. A skill for anti-money laundering. There are so many use cases once you understand the pattern.

Now imagine you are a marketer, a salesperson, an accountant, a chemist, a compliance person, or a specialist in any field. In your field, there are books. There are manuals. There are technical PDFs, white papers, product documents, research papers, training guides, and internal notes.

That is where book-to-skill becomes interesting.

You can take material you have the right to use, such as books you bought, public documents, open materials, company documentation, or technical manuals, and convert that knowledge into a skill. Instead of dumping a giant PDF into a chat every time, the skill becomes structured reference material that can be invoked when needed.

For example, let us say you are an anti-money laundering expert. You could take a large AML training manual, convert it into a skill, and suddenly you have a reusable knowledge layer around that material. You can invoke it when you are working, asking questions, drafting, reviewing, or thinking through a compliance issue.

You could do this with one book. You could do it with ten books. You could do it with a collection of manuals and papers. The point is not that the AI magically becomes perfect. The point is that you are giving it a structured, domain-specific body of knowledge to work from.

That is powerful.

The project is open source, available on GitHub, and it had about 28.7k stars at the time I wrote this. Try it with a book, manual, or PDF you know well and see how it changes the way you work with an AI assistant.

Useful link:

• book-to-skill on GitHub