July 8, 2026

KaKeKiKoKu Studio 1.14: a graph on a budget

The release where relationships learn the same discipline as everything else in Studio — and where Studio learns to teach itself.

KaKeKiKoKu Studio 1.12 made relationships visible. You could finally see the arrows, create them from anywhere, move their endpoints, watch your project become a real graph. And almost immediately, the obvious next question showed up: how connected is too connected?

A budget for the graph

It started, honestly, with a question from an AI. We were reviewing the limits system when it asked me: have you considered that in a large project, one resource can end up referenced by hundreds of others? I had not. I had spent months making relationships easier to create, and no time at all deciding what happens when creating them succeeds too well.

If you have read the bounded blocks post, you already know how Studio answers this kind of question. Not with a bigger buffer. With a boundary that means something.

So in 1.14, relationships have budgets. Each resource holds up to 12 outgoing relationships, each description up to 200 characters, all of it inside a character budget of its own. Twelve well-described links tell an AI exactly how your knowledge connects. Forty vague ones are noise wearing a graph costume. And the thirteenth link is the same canary as the 4,001st character of a section: the resource is covering too much, and splitting it is the better move.

Here is the part I like most: the budget gave space back instead of taking it. The prose cap for file sidecars and entities rises to 52,000 characters, and the counter in the path bar now measures only what you write — relation lines no longer eat your writing space. Next to it, two small chips track the graph: the outgoing chip shows how many links you hold and a thin bar shows how much of their budget you have spent — write lavish descriptions and fewer links fit, economize and more do. If you use AI coding tools, you already know this meter; it works like the token gauges you watch every day.

And the popular resources — the ones everything points at? Being referenced has no cap; you cannot control who links to you, and you should not have to. But when your AI reads a heavily-referenced resource, Studio now curates the view: the most descriptive backlinks first, then a simple note of how many more exist. The web interface always shows everything. The AI gets exactly what it can use. However connected your project becomes, every read stays lean, predictable, and worth its tokens.

None of this is a technical shortage, and by now you know I would tell you if it were. It is the same bet Studio has made since the beginning, applied to the last place it was missing: the graph itself.

Studio teaches itself

The other half of 1.14 is the KaKeKiKoKu Studio Learning Lab. Studio's whole thesis is that knowledge belongs inside the project, next to the files it explains. It would have been a little embarrassing to teach that thesis with a PDF. So the tutorial is not a document about KaKeKiKoKu Studio: it is a real Studio project — real folders, real context, real entities, real relations — and you walk through it inside Studio itself.

The moment you choose your working directory, Studio offers to place the Learning Lab there, in your language. One click: the project lands in your workspace, becomes visible to your AI, and opens ready to explore. If you skip the offer, a discreet Interactive tutorial link on the Welcome screen keeps the door open.

Inside, ten short stops cover the first mile: files with memory of their own, folders that orient everything living inside them, knowledge that belongs to no single file, relations you can see and touch, synchronization, images as knowledge, and what changes when an AI enters a project with a map instead of a blindfold. At the end there is a playground with four guided exercises — edit a context, create a file with its memory, connect two pieces, break something small and fix it. The exercises only touch what you create, so the tour stays clean.

It is deliberately not the manual. The manual explains every button; the Learning Lab teaches you to walk, and tells you where to go when you want depth. Delete it whenever you like — it is just another folder in your workspace. Although if Studio did its job, by then you will have stopped thinking of folders as just folders.

Small things that were missing

Since 1.12, opening a project quietly made it visible to your AI — every time, automatically. In 1.14 that is no longer a given: you choose which open projects your AI can see. Which introduced a small honesty problem. If a project is open but hidden and your AI cannot find it, nothing on screen explains why; it just looks like Studio failing to connect. So now an open project that sits outside your AI’s reach carries a small mark next to its name, and one click takes you to where you can share it. A quiet indicator for a quiet decision.

A couple of other rough edges got sanded down. The resource browser used to fan open every neighbour of whatever you selected, chasing each relation outward — helpful in theory, noise in practice. Now only the selected resource expands, and the rest wait until you ask.

None of 1.14 is loud. A boundary for the graph, a tutorial that lives where the knowledge does, and a few edges made honest. That is usually how Studio moves: one deliberate decision at a time, in the direction it has always pointed.