Student guide

Studium plays a lesson one step at a time and chooses what comes next from your answers: a hint, a detour through a prerequisite, or the next idea. An AI assistant can explain a tool, prepare a calculation or draw a graph for you. Your prediction, your reasoning and your conclusion stay yours. This page covers what to do in a lesson, how to use the assistant without handing it the work, and how to stop and resume.

The learning loop: a scene is presented, the learner answers, Studium gives feedback and chooses the next step.

Before your first session

Ask your teacher for the course lesson (a lesson.json with its assets), the Studium version and which experiments you may run. A PDF alone is not a playable lesson. To try right away, use the atomistique-l1 lesson shipped in fixtures/.

  1. Follow the quickstart.
  2. Choose where to play: the VS Code lesson pane, which needs no AI account, or an assistant such as Codex, Claude Code, Copilot or OpenCode over MCP (harnesses).
  3. Play the first lesson tutorial.

If your teacher gives you a portable activity (index.html), open it in a browser: nothing to install.

Class telemetry is off unless you opt in. While you try things out, use a test learner id rather than your real name; your teacher explains the rules of a class pilot before any collection. To learn without a teacher, from your own sources, see learn with your assistant.

What to do in each scene

Scene Your part
Introduction, concept Read, look at the figures, then continue
Quiz Reason first, then choose; ask for a hint if you need one
Confidence check Give your own confidence, from 0 to 4
Reproduction Read the protocol, predict, then examine results that were really run
Spot the flaw, summary audit, AI critique Tie each objection to a fact or a piece of evidence
Explain back, comparison Write your own formulation and its limits
Discussion Build the argument over several turns
Prompt practice, communication Draft, then improve your own proposal

Not every course uses every type; see scene types. The pane shows the right controls for each scene. In a chat, each step tells you how to answer: a letter, a number, or your own words.

The VS Code lesson pane showing a quiz question with three answer choices and a continue button.

If you are stuck

Try your own answer first, then use the help ladder in order: reread the step, ask for a hint, ask for an explanation, and read the reference. An explanation becomes available after you have tried an answer and hands your question to the assistant; the reference comes with the verdict, after your answer. Rereading costs nothing, while a hint reduces the mastery credit of a correct answer and brings its review sooner. See language and supports for the other kinds of help you can request.

See your concept map

The concept map shows each concept's state, what it needs first, the path you travelled and the assessed answers behind it. Open it in the lesson panel or pane, in forms, or ask in the conversation for the map returned by get_workflow_graph. On a concept you have reached, you can say that you know it or are unsure: Studium asks one or two check questions and reports their outcome; the claim alone never raises mastery or skips a scene. See Concept map.

Use the assistant without giving it the activity

Requests that keep you in charge:

Give me a hint about the approach, then wait for my attempt.

Show these data with their units. Ask me first what I predict when the threshold changes. Keep the reference visible next to the variant.

Here is my explanation: […]. Ask me one question that tests its limit, without replacing my text before I submit it.

Explain the last adaptation with the Studium tool. Separate what it reports from your own interpretation.

A hint lowers the mastery credit of your eventual answer; it is still worth asking when you are stuck. A detour sends you to a prerequisite and then back. Mastery figures are a learning aid, not an official grade or a diagnosis. See the learner model.

Experiment with an article or a capsule

Follow the agentic experiment tutorial: pick a question, predict an effect, run a reference, change one factor and judge the difference. Record your prediction before seeing any result in the experiment journal.

The assistant may search documentation or read a skill, but it must say what it actually consulted and ran. Check units, data, edge cases and provenance: a graph an AI produced can be wrong. Keep outputs in your experiment folder. A capsule marked type-checked is not a successful run. If a dependency is missing or a calculation fails, keep that failure; never ask the assistant to make up a result. In VS Code, a reproduction asks you to confirm the exact command before anything runs.

Language, representations and reports

Ask for another language at any time ("Continue in French"); your answers and the original sources are kept as they are. You can ask for a readable equation, a diagram, a chemical structure or a plain-text explanation. See language and supports.

To keep a record of your own actions, see the learner report and the activity-to-report tutorial. You choose what goes in; producing a report does not send it to your teacher.

Stop and resume

  • Pause. Say "stop here"; the assistant calls end_session, which closes the session and keeps your progress. Closing a window alone does not erase anything.
  • Resume. Start the same lesson with the same learner id: Studium resumes at the first unfinished step. A finished lesson starts again from the beginning as a retake.
  • Several places. The VS Code pane and an assistant keep separate progress. Two assistants share it when they use the same data folder and learner id.
  • Erase. Ask the assistant to forget your data for a lesson, or all your data (forget_session); it then tells you what stays on disk. See learn with your assistant. In VS Code, Studium: Forget my data erases what the extension keeps, not an assistant's data. Experiment folders and the assistant's own history follow their own retention rules.

Spaced reviews bring concepts back over the following days: see review over time. If something goes wrong, see troubleshooting and send your teacher only a small, anonymised example. Privacy details: privacy and limits.

View the source of this page on GitHub