Octatrack MKII guide: Parts, patterns and an AI helper

You turn up the reverb on a track in one pattern and move to the next pattern. The reverb has changed there too. You go back to fix it and start wondering what else you have moved without meaning to. For many Octatrack MKII owners, this is where the machine stops feeling like a sampler and starts feeling like a puzzle.

The answer is usually Parts. We built the Pricing Logic Octatrack MKII guide to explain this kind of thing in plain language, with the relevant manual pages one click away. It also includes an experimental AI helper for questions that don't match a heading.

Parts and patterns, in plain terms

A pattern holds your sequence: the trigs and the parameter locks you placed on them. A Part holds the sound setup the tracks start from: the machine, the default sample, the effects and the base parameter values. Several patterns can share the same Part.

When you change a machine, swap a default sample, adjust an effect or move a base parameter, you are editing the Part. Every pattern that shares that Part inherits the change. Editing trigs or locks is different, because that data belongs to the pattern you are working on.

Saving catches people out in a similar way. Saving your project state does not automatically save the audio in a recorder buffer as a file. If you have captured a loop you want to keep, saving it as a file is a separate job. The guide covers both, and links to the manual pages that describe each one.

What the guide covers

The guide has 24 authored topics based on Elektron's OS1.40A manual. They cover Parts, patterns and scenes, Flex vs Static machines, recording, sample files and sample slots, saving, routing and sequencing. Every topic shows links to the manual pages it draws on, so you can check the source yourself. The guide is independent of Elektron.

The notes are ordinary readable pages. They work without JavaScript and need no account. You don't download a model. Your browser receives lightweight HTML, CSS and JavaScript, and the model stays on our server.

Asking the question you actually have

Manuals are organised by feature. Problems are not. "Why did editing one pattern change another?" touches Parts, patterns and machine settings, which live in different chapters. The AI guide lets you ask that question as you would say it. It selects relevant notes from the guide and writes a short answer based on them.

The source notes stay next to the generated answer. You can read what the model was working from and follow the manual links from there. This helps most with ideas that connect to each other: how scenes relate to Parts, why a sample slot differs from a sample file, when Flex or Static suits a job. Explaining one of these properly often means drawing on two or three sections of the manual at the same time.

The guide runs from a phone, with no local GPU needed. If your setup is a desk, an Octatrack and whatever is in your pocket, that is enough.

What the Explain further button does

When an answer has finished, you can click Explain further. This asks the model for a deeper explanation of your original question, using the same selected notes.

The model receives your original question and the selected notes, rather than the full conversation or its previous answer. Each click is a fresh, more-detailed explanation of the same question and the same source material. If you want to ask about something else, ask it as a new question.

The transcript on the page is held in memory while you use it. If you want a copy, you can save it yourself.

How it works, from your phone to a GPU and back

When you ask a question, it travels from your phone or laptop to the public website, which is hosted on Vercel. From there, a Cloudflare Worker and Tunnel pass it to a private, authenticated guide backend. The backend picks the notes relevant to your question and sends them with the question to Qwen3.8-27B, running on an RTX 5090 in Pricing Logic's server. The answer streams back to your screen as it is written. Your browser never sees the raw model API or its credentials. The page shows which model is answering and where it is hosted.

Your question and the selected notes do travel over the internet to that server. We have not audited how long each upstream service retains data, so leave out anything you would not want to send.

The AI part needs a connection. It may sometimes be paused or unavailable, but the notes stay readable either way.

Measured speed and honest limits

On 7 October 2026 we sent five questions one after another through the public HTTPS route. All five came from a single development connection, and the service was already warm. All five answers completed normally. The median time to first text was 3.1 seconds, and the median time to a full answer was 3.2 seconds. Full answers took between 1.8 and 7.9 seconds, and they were 66 to 122 output tokens long. These times include network transit.

That is five requests. We ran no concurrency test and took no measurements on a physical phone. Treat these numbers as a smoke check of a warm service. They are not a benchmark or a promise about the speed you will see.

On accuracy, the AI guide is experimental. In an earlier evaluation of 30 questions, 7 answers had material errors or omissions. Since then, we have made one narrow live fix to how Flex and Static notes are retrieved. Even so, the guide is not a replacement for the manual or for an experienced user. That is why the sources sit beside every answer. Check them before you rely on an answer, especially before any step you can't undo.

The scope is narrow on purpose. The guide is tied to OS1.40A, not to whatever firmware is newest. It gives no recovery or formatting advice, and it can't diagnose hardware faults. It cannot see or control your instrument. It only knows what you type.

During the beta, answers are capped at 512 output tokens. The guide allows up to 100 model requests per UTC day across all users, one at a time, with a small queue behind the active request.

If a pattern has just changed behind your back, start with the Parts topic in the guide, read the linked manual pages, then ask your question in your own words.