AI Statement

You have probably had this argument already, or watched somebody else have it. Someone posts a track, someone asks whether a machine made it, and the thread runs for days. It is a real argument, it is not going away, and if you are thinking about buying a plugin with a model in it you are entitled to know where that plugin stands. So this page says what is in Bacara, how it got there, and what I think about the rest of it. It is longer than a product page should be. The subject earns it.

The objection is a good one. Musicians are angry about AI and the reasons hold up. Work was taken without asking, to train models that now compete with the people it came from, and a great deal of what comes out is sold as music without anyone having made any. If that is where you are starting from, I am not going to try to talk you out of it, and this is not a defence brief. There is a model in Bacara. I built Bacara with Claude. Both are true, both are below in detail, and you can make your own mind up at the end.

A note on who "I" am, since this page leans on the word. Bonboa is me, Joris Röling, on my own. Friends test Bacara, argue with me about it and write style documents for it, and they are credited where their work shows.

I am not a player. There is system in music, and describing that system has been a preoccupation of mine for a long time. What started Bacara was irritation with the levers a DAW hands you. Turn the cutoff up. Edit the piano roll until it is right. They are practical and they are not how I think about music. I would rather say density, anger, rest, and have the machine work out what that means on this bar, on this instrument. Parametric Directing, if it needs a name. The styles in Bacara are a first piece of it: a style document says how dense a part is and where it leaves gaps, and the generator works out the notes. The rest is not built. Bacara is my handicap turned into a product.

Here is what is actually in the box.

The generator. A character-level language model of order 18. It reads a pattern one step at a time and predicts the next step, the way a text model predicts the next letter. It is small enough to run on a microcontroller.

What stays on your computer. The generator reads its model files from disk and has no network code. Your patterns and projects never leave your machine. Bacara goes online for one thing, the licence and trial check. That sends your serial number, the name and email you registered with, your computer's name and device ID, the Bacara version and the trial's generation count.

Notes, not audio. Bacara writes notes and sends them to your instruments. It has a demo sound so you can hear a pattern before you connect a synth. That sound is calculated on the spot inside Bacara: no samples, and no model behind it.

Your instruments stay as they are. Bacara does not change the firmware of anything it plays. It sends MIDI to your synths and drum machines, the same as any sequencer, and they keep running the software their makers shipped.

Styles. Next to the generator sits the style system. A style is a document that describes how a part behaves: how dense it is, where it syncopates, where it leaves gaps. Bacara reads the document and generates new material from it each time you roll. I wrote the style documents with Claude.

What the generator learned from. Step patterns: note, accent, slide and gate on a grid. Acid lines, drum patterns, and patterns I wrote for styles I could not find elsewhere. I wrote those with Claude as well, and the chord generator learned from them alone. No recordings went in.

What you make is yours. You can use and sell every pattern Bacara writes. I claim nothing on it and I never see it.

How it got built. Bacara started in May 2021 as a prototype. The generator came a year after that, 167 lines of JavaScript that read a pattern character by character at order 18, and the plugin you can buy today still generates at order 18. The C++ began in 2024. All of that was written by hand, and it was on stage before a model ever touched the code, in generative performances with Richie Hawtin and at IRCAM in Paris. Since early 2026 a lot of the work has been done with Claude, Anthropic's model: code, the style documents, and the curated patterns the generator learned from. I review and test what it writes, and when something breaks that is on me. That review is where the usual line runs between AI-assisted coding and vibe coding.

What I cannot tell you. Anthropic does not publish what Claude was trained on. I can list what went into Bacara's own model. I cannot list what went into the model that helped me write it, and some of that reached Bacara's generator through the patterns I wrote with Claude. I would rather tell you that than leave you to find it out.

What might change. I have asked it directly. Ask a model what it drew on for a particular piece of work and the answer comes back as "training", which is true and no use to anyone. Regulation may force that to change and I hope it does. Nearly every style that ships with Bacara carries a note of what it was built from, and the Credits panel in Bacara shows it. The day there is a better answer, there is somewhere to put it, and this page gets it.

Music is made from other music. That part is older than any of this. Producers learn by copying a line they liked. A rhythm crosses from one genre into another and stays there for thirty years. The music Bacara is for was made that way from the start. Every generation had a machine that made it easier, and every generation was told the machine was cheating. Roland sold the 303 as a box that would play the basslines so you would not have to hire a bass player. It failed, went secondhand for next to nothing, and people who could not afford anything else turned the knobs the wrong way and built a genre out of it. How hard a tool is to operate has never said much about what comes out of it.

Bacara writes patterns. You make the music. A pattern is a starting point. You choose the scale and the style, you set the deviation lanes, you decide which roll to keep and which to throw away, and you play it into whatever makes the sound. Roll it again and it gives you something else. Nothing leaves Bacara finished. The hand and the ear at the end are yours.

What Bacara is for. Some products take a sentence and return a finished track, and a lot of that is on the streaming services now. Bacara does not do that and is not built to. It writes MIDI notes into your sequencer and stops there. Judge it the way you would judge a sequencer, because that is what it is. It is for someone who already has a session open and needs a part that keeps moving.

So, is it all right to use? I think so, and this is the plain version of why. Bacara will not write your track. It hands you a pattern, some worth keeping and most of them not, and you throw the rest away, which is the deal every sequencer has ever offered. Whatever reaches a listener got there because you decided it should. If somebody asks whether a machine made it, the answer is the one it has always been: a machine helped, and you made it.