Use any AI (Claude Code, Codex, Cursor, Claude or ChatGPT). It connects to Kater through the Kater MCP.

Your agent follows rules and context in your repository’s ai_tools/ folder.

You can read every file in it, change it, copy it or add your own. This section tells you what is in the folder, how your agent uses it, and how you make it yours.

How Kater is designed

Kater’s AI tools are inspired by the BMAD Method, an open framework for AI-driven development. BMAD gives an agent named roles, workflows in steps and checkpoints, all in plain files that the team owns. Kater takes the same approach:

• We do not own the harness. You pick the agent, and you can change it at any time. Kater connects to it and gives it what it needs to work well on your data.
• We give you the toolkit. Design your own skills and workflows. Write down your business context and who you are. Kater sends it to your agent, so your agent learns about you.
• You have full control. Every agent, skill and workflow is a file you can read and change. Nothing your agent follows is hidden in a prompt that you cannot see.
• You start from what works. Out of the box, you get the set we tried and tested for AI-driven analytics. Most of the work is built for you on the first day.

What you get out of the box

When your platform admin first initializes a Kater project, the ai_tools/ folder is scaffolded. Each file in it has the tag kater-default.

The defaults are yours from the first day. Change one, and your change stays. Kater never writes over a file that is already there.

What your agent reads, and what it writes

Your agent does not load the whole folder. It reads a few files when it connects, a few more when a task starts, and only what the request needs after that. The color of each tag tells you who can change the file.

• When a conversation starts. The Kater MCP gives your agent Kater’s rules and a list of files to read: your organisation’s rules.md, the company’s about.md and your own about.md. Your agent reads them before it answers.
• Before it works. It reads the connection’s connection_context.md, the AI tools’ config.yaml, and the guide in guides/ for each object it changes.
• For the request. It loads one agent, one workflow step at a time, and the skills that step names.
• What it writes. Plans, drafts and notes go in your brainstorm folder with no approval. A lasting fact about you goes in your about.md. Anything else waits for your yes.

How it runs follows one request from start to end.

The folder

about.md                          About the company
rules.md                          Your organisation's own rules
guides/                           Kater's guides, one for each object
connections/<connection>/
  connection_context.md           Business terms and data limits
  ai_tools/
    README.md                     What each object is
    config.yaml                   Folders and the run mode
    engine.md                     How a workflow runs
    agents/                       conductor.md, planner.md and four more
    skills/                       brief-me.md, check.md and 26 more
    workflows/                    deep-analysis/, monday-brief/ and three more
  users/<you>/
    about.md                      About you
    brainstorm/                   Your agent's plans, drafts and notes
    ai_tools/
      agents/                     Your own agents
      skills/                     Your own skills
      workflows/                  Your own workflows

Your own folder starts empty. Kater makes each part the first time something is written there.

Who can change what

Access follows the same grants as the rest of your content. These are the usual holders.

Everyone can read the shared files. Your own folder, users/<you>/, is private: only you and the admins can open it. See Access.

Next steps