AI tutor for any subject, at your level

Requests it is built for

These are the kinds of requests the product is designed around, written the way people type them. Each is an example of what to type, not a quote from anyone.

• "My son is stuck on fractions and I cannot explain them" - a parent who needs to sit with her child on Thursday night and actually help.
• "I have a Google interview in six days" - a candidate who wants the questions and scenarios they will face, not generic advice.
• "Just the turbocharger. Nothing else about the engine." - a mechanic who needs one subsystem explained properly, once.
• "Get me through the cardiology chapter by Thursday" - a student with one chapter between them and the exam.
• "I have to learn Rust by Friday" - a programmer who needs the concept, not a 40-hour series.
• "Show me how to use the new scanner before Monday" - a dentist with new equipment and a first patient booked.

It fixes your level before it teaches

Before anything is built, the course fixes what you already know, so nothing you understand gets explained to you again. There are four levels, and the level changes how the course is pitched, never what it costs.

Beginner assumes nothing and builds every idea from first principles. Intermediate assumes the vocabulary and skips the basics. Advanced assumes the formalism and goes straight at the hard parts. Expert assumes the literature - no recap, no analogies.

One concept, a whole subject, or your own material

Three tracks. The composer suggests one from how you phrase the request, and you can change it.

• One concept: one idea, from first principles to the point where you can use it. Defaults to 10 minutes - eigenvectors in one evening.
• A whole subject: modules in order, start to finish. Defaults to 60 minutes and scales to 30 hours.
• Your content: a chapter, a paper, a contract, a deck, a spec, a photo of a worksheet. Taught back in plain words. Defaults to 15 minutes.

Length follows the subject

Runtime is yours to set, from 30 seconds to 30 hours. Twelve minutes if twelve minutes is the truth. A longer request plans more modules, each a few minutes, rather than a few modules stretched to fill the time.

Minutes are the only meter, at any level: the Free plan includes 30 minutes a month, watermarked and with no card needed, so you can make something real before paying anything.

Where the module list comes from

No step asks you to approve a module list before you commit. The runtime you pick is the lever that decides the shape: no module may run longer than 30 minutes, so a longer runtime is split into more modules rather than longer ones, and the deterministic plan for a 30-hour course is 225 modules of 8 minutes.

The chapter titles themselves are decided when the build starts, by a model that reads the whole brief - your own words, every answer you tapped, the track, the level, the runtime, and each file you attached with the line you wrote on it. If that model cannot be reached or answers unusably, the deterministic plan is used instead. Either way the plan lands on the build card in your library, and the event log on that build names which of the two decided it. If particular chapters matter to you, say so in the notes box in your own words.

Handing over a chapter or a worksheet

A request can carry reference files, and the Files step is where they go. It is optional - the default answer is no and it costs one tap. The picker offers PNG, JPEG, WebP and GIF images, PDF, plain text and Markdown, and nothing else: there is no DOCX, PPTX or EPUB path, and a file dragged past the picker is refused with the same sentence.

You can attach up to 5 files, each with your own one-line note. An image is resized in your browser and uploaded to storage only our build worker can read. A PDF or a text file is never uploaded - its text is read in your browser, and only that text travels with your request, where the model that plans your course reads it.

Each file has to be under 10 MB, at most 20000 characters are read out of a PDF or a text file, and what you attach is material to work from rather than an instruction that must be obeyed: the pipeline decides whether each one fits the video it is building.

The "Your content" track asks for the document by name, and the name is what travels into the brief: that picker reads nothing and sends no bytes. If the text of the document itself has to reach the build, attach the document at the Files step.

Keeping it, and sharing it

Every finished build gets a watch page of its own. It is private to begin with - you, and the team on your account - and you can open it to people you name by email address, or make it public, in which case anyone holding the link can watch it and only then can it be embedded on another site or posted to a social network.

Asking for a change afterwards

A change to a finished video is a NEW BUILD, not an edit inside the file you already have. You type the change in plain words, a confirm strip names the cost before anything is spent - the full runtime of that video in minutes, so a 30-hour course is 1800 minutes again - and the new build appears in your library beside the first one.

Nothing in the product edits inside an existing file, so there is no way to ask for one stretch of a video to change and leave the rest alone. The build you have is kept exactly as it is, with its own event log, and the new one is a build of its own.

Early build: an update request spends real minutes and hands back another sample file, not your rendered course. The video you have now is kept exactly as it is, and we will tell you the day real rendering begins.

What this early build does today

ZUQ is an early build, and the part of it that is live is the part that costs money. A request is saved, minutes really are debited, the build runs and its chapter list is decided - and then it hands back a sample file rather than your rendered video. We will say so on this page the day real rendering begins.