alpha

For developers

Translation of the project's DEVELOPERS.md; the Czech original is authoritative. The project's documentation, task files, issues and commits are in Czech — Claude Code reads them for you, and you can ask it in English. Discord is Czech-speaking too, but English works as well.

Shipard is an open-source accounting system (MIT licence) whose code is written by AI. People decide on the architecture, the rules and the assignments — and read every diff.

Want to run it right away? Mac or Windows → one-command installation · Linux → manual installation

Shipard is in alpha: the main things run, quite a bit is still changing. What it can and cannot do today is described in the README.

Why again, and why this way

The new Shipard is a rewrite of an older system of the same name. Behind it are twenty years of building similar systems — we know what we would do differently today, so we are doing it again, from the ground up.

One rule has applied from day one: the code is written by AI. There is not a single line of code in the whole system written by a human. People direct the architecture, the rules and the assignments, and check the result.

It is not "write me an accounting system". It is a controlled development process with locked decisions, detailed documentation of principles and rules that bind the AI. We do it this way because it pays off: with this method, AI-written code has a better ratio of cost to result than hand-written code. Manual programming is over for us.

How we do it

  1. Design. The problem is discussed with Claude in chat. The decisions are numbered and locked in a GitHub issue before any assignment is written.
  2. Assignment. The decisions become a task in tasks/: what to read, what to do and how to tell that it is done.
  3. Implementation. Claude Code writes the code and tests according to the task — usually a different model than the one the design was made with.
  4. Review. The finished work goes back to chat and is gone through in full, against the assignment and against the surrounding code.
  5. A person reads the diff and decides what gets pushed. AI pushes nothing on its own initiative.

The code thus gets more attention than if only people wrote it. In detail: Developing with Claude.

In the repository history you will also find commits without Claude's signature. A person commits — the code in them was still written by AI.

You don't have to read the documentation first

The rules and documentation are extensive: CLAUDE.md, dozens of documents in docs/ and over three hundred task files in tasks/. They are written mainly for the AI, which reads them with every piece of work. You don't have to read them beforehand.

Want to know how something works? Ask Claude Code in your checkout. It answers from the code and the documentation, not off the top of its head.

Core and modules

Shipard is a core: users and sign-in, a module system, a server-driven user interface and, on top of that, the usual areas — documents, accounting, VAT, bank, assets.

The main thing is extensibility. When you are missing something:

  • Is it useful to everyone? It belongs in the core — open an issue or send a pull request.
  • Does one company need it? Make it a private module outside the main repository. A module adds tables, lists, forms and its own logic; custom frontend components not yet (docs/modules.md, chapter 10).

In both cases you write an assignment, not code. And you are responsible for the result.

How Shipard is developed

Shipard runs on Linux (Ubuntu) as a server — nginx, PHP, MariaDB. It is therefore developed remotely (remote development): the code does not live on the disk of your Mac or Windows machine, but on a Linux machine you connect to. Claude Code runs directly on it and you give it work from the terminal.

That machine can be:

  • a virtual machine inside your computer — Multipass on a Mac, WSL on Windows. It is the same server, just inside your computer. Enough for trying things out and for everyday work.
  • your own Linux server — for a team and serious development: it runs all the time, holds more data sources and more people can sign in to it.

What you will need

  • A computer with about 20 GB of disk space and 4 GB of memory for the virtual machine, or a server with Ubuntu 26.04.
  • A paid Claude account (Pro, Max, Team or Enterprise) for Claude Code. Just running Shipard does not need one.
  • A GitHub account, once you want to send your work.

You don't need to know how to program. You need to be able to describe precisely what you want and be willing to check the result.

Where next

I want to run Shipard

  1. The environment — on a Mac or on Windows, or on a Linux server.
  2. First steps and daily work — the developer dashboard, your first data source, what to do after git pull.

I want to fix or add something

  1. Claude Code for beginners — installation, permissions, how to assign and check work.
  2. Developing with Claude — roles, the process from issue to verification, rules.
  3. What an assignment looks like — and over three hundred finished ones as examples.
  4. Fork and pull request — how to send your work.

I want my own module

The module system describes the structure of a module and, in chapter 10, modules outside the main repository. The fastest way: describe to Claude Code what the module should do and have it propose the tables and forms.

I want Shipard for my country

We are starting with Czech accounting; the goal is a system for companies anywhere in the EU. Another country is not a translation of the interface but a country module built on the core — the intent is described in the roadmap (section "Za horizontem", beyond the horizon). Do you know the accounting and tax rules of your country? Get in touch.

I'm looking for how something is done

Where to ask

A quick question — "is this intended, or a bug?", "how is X done?" — belongs on Discord. There's only a handful of us there and we answer quickly.

Report bugs and ideas as issues on GitHub. Reports are public — no real names, amounts or account numbers belong in them.