Gamemasters · AI Wizard

Generate a Whole Campaign with the Wizard

Stay at campaign scope and let the AI Wizard draft a full tree of paths, nodes, and vocabulary in one multi-segment run.

Requires PatronRole: Gamemaster (campaign owner)

Verified against app build 2f427ecbd · 2026-08-25

You can scaffold an entire campaign in one pass. The wizard generates structure inside a campaign you’ve already created — so make the campaign shell first (here’s how), then let the wizard fill it with paths, nodes, and vocabulary.

Subscription

Requires Patron or higher.

Like all AI generation, this needs the Patron tier or higher. A full-campaign run is the most token-hungry flow, so a Patron’s 200K monthly grant can run out mid-run — already-finished segments stay committable.

There's no single “Generate campaign” button

Whole-campaign generation is a starter card: Sketch the learning paths for the skeleton, or Build the full course for the broadest single pass. The server expands either into a multi-segment run (paths → nodes ‖ vocabulary → coaching). Staging it this way always leaves room to set the counts and add guidance before you send.

Scaffold the campaign

  1. Open the wizard at Campaign scope

    In the Builder, open the AI Wizard and leave the scope button alone — at campaign scope it reads your framing concept, and the session is labeled Campaign Wizard. The wizard greets you with the framing concept and offers a card for each thing it can make here.

    The AI Wizard at campaign scope, its scope button reading the campaign's framing concept
    At campaign scope the scope button reads your framing concept, not the word “Campaign”.verified @ 2f427ecbd
  2. (Recommended) Ask the wizard to plan first

    Ask the wizard to plan describes what you want in plain words and gets back a numbered list of proposed steps instead of content. The wizard reads what is already in the campaign, says what it found, shows which of your essential questions are already covered, and proposes changes.

    Every step is editable: rewrite a directive to say what you actually meant, or ✕ it out of the plan. Below them sit Step back, Dismiss, and Approve, which carries the number of steps you kept. Asking a question in the chat field leaves the plan — and your edits — intact.

    The proposed structure plan at the approval gate, listing editable per-path directives above the What I found diagnosis
    Nothing is written until you approve. Each step can be rewritten or dropped first.verified @ 2f427ecbd

    Planning is not free

    A plan costs tokens like any other request — it is cheaper than a generation, not exempt from your quota. What it buys you is the chance to see and edit the steps before the expensive half runs, so a wrong assumption costs a plan instead of a whole campaign.

  3. Or go straight to a draft

    Tap Sketch the learning paths for the path skeleton, or Build the full course for paths, objectives and their content in one pass. Unlike the plan route, this one goes straight to a reviewable draft.

    Staging the card reveals the sentence that sizes the run — “I desire 3 paths, with 4 nodes each. Per node, align 2 new or reused vocabulary terms.” Every bold value is a pill you tap to change, and those numbers are what decide how long the run takes and how much of your quota it spends. Guidance below is optional.

    The staged Sketch the learning paths card above the fill-in-the-blanks sentence with tappable count pills
    The pills size the run — set them before you send.verified @ 2f427ecbd
  4. Tune Context & settings (optional)

    The tune icon at the bottom of the composer (Context & settings) opens the Context Wizard. Choose which campaign elements feed the prompt (framing concept, vocabulary, story settings) and set the balance — for example “Balanced,” “Learning First,” or “Story Forward.” A dot on the icon means the session is running on non-default settings. The paperclip beside it attaches sources for the wizard to ground its work in.

  5. Watch the segments, then review

    While the run is in flight, per-segment checkpoint tiles (Paths, Nodes, Vocabulary, Coaching) appear as each one completes, and they are the one place that reports what a segment actually cost. If a segment fails, only it shows a retry chip — the rest still land.

    Once the run settles those tiles give way to the durable record: a coaching summary, Next steps, and a changelog of what was created.

    The settled review state with the pedagogical rationale expanded above the changelog and next steps
    After the run settles, the coaching summary and changelog are what remain.verified @ 2f427ecbd
  6. Commit the whole structure

    Use Give Feedback to revise or Commit to persist the entire tree in one transaction (Undo rolls it back). The post-commit card shows Saved to <campaign> with count chips. From there, drill into a node and run object-level generation to fill its investigations, questions, and reports.

Mid-run stops

  • Quota exhausted mid-run: “You are out of AI tokens. Your next 200K tokens arrive on your renewal date …” Completed segments remain committable.
  • Coaching deferred: if some content segments didn’t succeed, you’ll see “Coaching deferred — retry with full budget.”
  • Segment failure: per-segment retry chips resume only the failed segment.