Quick Start

Get KaKeKiKoKu Studio running with your AI in minutes.

What KaKeKiKoKu Studio does

KaKeKiKoKu Studio is built for AI coding agents, such as Claude Code, Codex, Antigravity and OpenCode, and for the projects you build with them, code or not.

KaKeKiKoKu Studio is the application: a knowledge system where you and your AI shape your project’s knowledge. It stores that knowledge in the Wuniq format, using files with the .wuniq extension. It keeps two layers side by side:

Your project’s knowledge — decisions, rules, intent, and relations, structured alongside your files: folder sidecars, file sidecars, and standalone entities. Knowledge lives inside the project, not in a separate database or in any AI provider.
Your files — the results where that knowledge shows: code, documents, research, anything.

When your AI opens a project, KaKeKiKoKu gives it the root context and a summarized resource map. The AI then reads the relevant folders, files, entities, and sections as needed; opening a project does not load all its knowledge into the conversation.

How both layers stay in sync:

Your AI works on your project as usual.
KaKeKiKoKu detects when project files change and helps route the knowledge update back through your AI.
You can see and edit your AI's project knowledge from the KaKeKiKoKu UI at any time. The sidecars and entities are yours to inspect or rewrite.

Decisions, constraints, rejected alternatives, and gotchas remain available across sessions. Keeping that knowledge accurate still requires the AI and you to review it when the project changes.

Prerequisites

1 KaKeKiKoKu Studio

Go to the Download page and pick the installer for your OS:

Windows

Click Get on the Microsoft Store page. After install, launch KaKeKiKoKu from the Start menu icon, or run wuniq-mcp.exe in a terminal.

Linux

Install it from the Snap Store:

sudo snap install wuniq

For projects on removable or mounted storage under /mnt, /media, or /run/media, also run sudo snap connect wuniq:removable-media.

KaKeKiKoKu opens its UI at http://localhost:5050 by default. If you see it, you're good.

Start KaKeKiKoKu before your AI client. This is required when you connect through an instruction file. Launch it from the Start menu (Windows) or the applications menu (Linux) and it stays up until you exit it, independent of any AI client.

This is the way we recommend using KaKeKiKoKu, for two reasons. It is yours, not your editor's — a knowledge base you can open, read and edit whenever you like, with or without an AI in front of it. And it makes your first message faster: your AI connects to a KaKeKiKoKu that is already running, instead of waiting for one to start.

The MCP connection can also start KaKeKiKoKu when the AI first runs a command that needs it. Discovery and local help do not start the application. If KaKeKiKoKu is already running, the MCP reuses it.

Working directory

2 Set the working directory

On the welcome screen, set the "Working directory" field — the base folder where you'll organize your projects (e.g. C:\Projects or ~/projects). Configure it before opening or creating your projects.

Why? AI coding clients usually work from a single folder or workspace. The working directory aligns KaKeKiKoKu with this model: you open your AI client there, and your projects are subdirectories inside it. This is an organizational base, not a security boundary: your AI client's permissions and the operating system still govern access to files.

Important: Don't create a KaKeKiKoKu project directly at the working directory root. Always use subdirectories (e.g. C:\Projects\MyApp, not C:\Projects). KaKeKiKoKu will prevent this with a clear error message.

Learn KaKeKiKoKu from the inside: right after you set the working directory, KaKeKiKoKu offers to place the KaKeKiKoKu Studio Learning Lab there — an interactive tutorial that is itself a real KaKeKiKoKu project, in your language, visible to your AI and opened with one click. Ten short stops and a playground with guided exercises cover the first mile; the user manual covers the depth. You can also open it any time from the Interactive tutorial link on the welcome screen.

Connect your AI

3 Connect your AI to KaKeKiKoKu Studio

To look around first: add one instruction file. Your AI reads it and connects to the running Studio through local HTTP. There is no MCP server configuration to set up in your editor. It is best for a small project and for learning what KaKeKiKoKu is.

We recommend AGENTS.md, shared by most clients, including recent Claude Code versions. If you already have Claude instructions or your editor uses another rule file, keep them and follow the loading options in the common guide. Add the block without replacing existing rules, then start a new conversation.

Get the instructions and follow the common connection guide →

HTTP connects when your AI makes its first connection request. Until then, Studio can show “No AI connected”, even with the instruction file in place. Start a conversation and ask your AI to list the available projects. With MCP, your client can connect in the background when it starts its configured MCP server.

Works with local AI tools. The AI needs a local shell or another tool able to make HTTP requests to Studio. A remote or sandboxed agent may need different permissions or a local execution environment.

Once you decide to keep using it, switch to MCP. Over HTTP your AI writes a shell command for every request, long responses can arrive trimmed, and the extra steps can use somewhat more tokens. With MCP your AI gets a dedicated KaKeKiKoKu tool: each command is a single tool call, instead of a shell command it writes out and your client may ask you to approve. Your AI client can also start Studio when it needs it, and MCP supports channels in clients that deliver them. You can start with the file and switch later by replacing its KaKeKiKoKu block with the MCP one; your project knowledge stays the same. Follow the MCP guide for your client.

Verify

4 Verify the connection

Open a project in KaKeKiKoKu (a subdirectory of your working directory) and mark it as available to AI in the project list. When your AI runs a command against the Knowledge Engine, such as open list, you should receive the actual list of available projects. Check that result rather than relying only on a connection indicator or a help response.

The startup instructions from step 3 ask your AI to run open list each session. Verify that it does; if it skips the instruction, ask explicitly: "Use the KaKeKiKoKu connection in your startup instructions to list the available projects."

Tip: keep KaKeKiKoKu on screen while you work. KaKeKiKoKu is a local web app, so AI clients with a built-in browser pane can display it right beside the chat — one screen, no window switching, and you get to watch that indicator light up as your AI reads and writes knowledge. Whether your client supports this, and how to set it up, is covered in its integration annex.

Sync modes: Automatic and Precise

KaKeKiKoKu has two sync modes, selected in Settings > Application:

Automatic — KaKeKiKoKu can send requests and change notifications through channels. This requires an MCP connection and a client that actually delivers those channels to the AI; a green connection indicator alone does not establish support. See the Claude Code guide for the supported setup and the Sync panel for the notification countdown.
Precise — KaKeKiKoKu detects changes in your project, surfaces them in its UI, and recommends when to sync. When you decide, type sync in your AI chat. Ask the AI to run the KaKeKiKoKu sync tool and review the reported issues. Use this mode for the connection through an instruction file and local HTTP. It does not require channels.

Precise mode may sound more manual, but in practice it feels natural with strong coding models. Assistants such as Codex often remember to synchronize as part of their normal workflow, and KaKeKiKoKu's clipboard flow for "Talk to AI" prompts keeps the interaction smooth. If you want to force a synchronization at any point, type sync in the AI chat. Automatic is more seamless when channels are available; Precise is not a second-class mode.

Default for new installs: Precise. Switch to Automatic only if your client supports channels and you launched it with the appropriate flag — KaKeKiKoKu asks for confirmation when you activate it.

You're not just a spectator. KaKeKiKoKu includes an editor. You can — and should — open any sidecar or entity and edit knowledge directly: refine wording, add rules, capture decisions, fix what your AI missed. This works in both modes.

Ready to go deeper?

Read the full documentation →