Authoring with AI

This is the heart of Tessera. You describe what you want in plain English; your assistant produces real pages, components, and scored quizzes; the live preview shows you the result seconds later. You never touch the code unless you want to.

The loop

1. Run pnpm dev <course> and keep the preview open in your browser.
2. Open your AI assistant in the workspace folder and describe the change you want.
3. Watch the preview update as the assistant writes. Every save hot-reloads.
4. React to what you see: "make the intro shorter", "that quiz needs a fourth question", "the callout should come before the video".
5. Repeat until it's right, then commit it.

The unit of work is a conversation, not a file. You stay in the role of author and editor: judging the result on screen, not inspecting how it was made.

What your assistant already knows

The complete technical authoring guide (AGENTS.md) ships inside Tessera itself, and a pointer file in the workspace root leads your assistant to it. Because the guide is versioned with the framework, the assistant always has current, correct instructions for the Tessera you're running: course structure, every built-in component, quiz wiring, tracking rules. You never paste documentation or explain the tool. If the assistant can read your workspace, it knows how to build in it.

What a course is made of

So your requests have the right vocabulary: a course is sections of lessons of pages. A page can hold anything a web page can, plus Tessera's built-ins: callouts, images, accordions, carousels, video and audio with captions, click-to-reveal modals, and the question types. And you're not limited to the built-ins: your assistant can create custom components for anything a page needs, with the same tracking as the rest. Page and section names, order, and titles all come from how the files are arranged, and the assistant manages that arrangement from your descriptions ("move the recap to the end of module one").

How to ask

• Point at your source material. The assistant can read documents you already have and turn them into pages and quiz questions.
• Name the course when the workspace has more than one.
• Describe outcomes, not implementations. "Learners should be able to spot the three unsafe behaviours" produces better material than a list of widgets to place.
• Iterate in small steps. One section at a time reviews faster than a whole course at once, and course corrections stay cheap.

When the built-ins aren't enough

Anything your assistant can build for the browser can go on a page: a timeline, a calculator, an interactive diagram, a game. And custom interactions aren't second-class; built with Tessera's hooks, they score, track, and report to the LMS exactly like the built-in question types. See Quizzes & interactions for what that unlocks.