Prepping a published adventure for AI.

What prep is actually for

An AI run with no prep drifts. It works from a fixed window of recent text, so once the early scenes, the NPC from three sessions ago, and the faction you crossed fall out of that window, it fills the gap with whatever sounds right. No clever prompt fixes that. It is the best-documented way AI games fall apart.

Prep fixes it structurally. The published adventure you import becomes a fixed, searchable source the AI reads from, and Familiar's Knowledge and Memory layer keeps campaign facts on hand between sessions. The AI is not remembering the story from a conversation. It is reading the prep you built. That is the difference between a tool that holds the thread and one that loses it.

This is real work, not a button. The payoff is that it is front-loaded: you structure the adventure once, and the AI references it every session after.

It does not drift, because it is not inventing the plot. It runs the encounters, NPCs, and rulings you laid into the journals, in the order they unfold.

The three artefacts to build

Most of a published adventure is already structured the way the AI needs. Your job is to land it in the right Foundry documents so it is searchable and linkable, then add the one thing the module does not give you: a short outline of what happens next at your table.

Build these in order, inside Foundry where the AI can reach them. Full-text search runs across every journal, character, scene, and item, so the rule is simple: if the AI needs to know it, it lives in the world, not in your head.

1. 1

   Story beats as journals

   Import the adventure so its scenes live as linked journal pages: the opening location, the first dungeon, the key encounters, each as its own page, in the order they unfold. These pages are what the AI reads when the party arrives somewhere, so the wording on the page is the wording it works from.

2. 2

   The cast as sheets and statblocks

   Pull every NPC and monster you will need from a compendium into real character sheets and statblocks before play, not mid-fight. A statblock the AI can read is one it can run: initiative, attacks, saves, conditions, death saves, all from real numbers rather than improvised ones.

3. 3

   A short session outline

   One page that says where the party is, the two or three scenes likely this session, and the encounters waiting. This is your runbook, not a script. It points the AI at the relevant journal pages and statblocks so it opens the session already oriented.

• Journals & Notes
• Characters & Items
• Knowledge & Memory
• Combat & AI

What goes in the session outline

The outline is the artefact people skip, and it is the highest-leverage hour of prep. It is the AI's dashboard: a quick read of the current state plus a map of the scenes ahead, so the model never reconstructs the campaign from scratch.

Keep it to a single page. It points at the journal pages for the full text rather than duplicating them, so the AI reads the detail on demand and the outline stays scannable. Familiar's persistent memory bank and its continuously-rewritten plot summary carry the load between games, so each outline starts from where the last one ended.

• Where we are: the party's location, the in-fiction time, and the immediate situation, in one or two lines. This is the cold open the AI recaps from.
• Pickup from last session: what was resolved, what was promised, who is hurt, and any open thread the AI should not lose. The memory bank holds this; the outline restates the load-bearing facts so nothing important sits only in the model's context.
• The scenes ahead: a short arc map of the likely next beats, each with more than one way it can go, so the AI follows the branch the players choose rather than steering toward a single scripted outcome.
• Combat readiness: a note that the statblocks for the likely fights are pulled and named, so the first encounter does not stall on a missing monster.
• A source index: pointers to the journal page behind each beat, so the AI fetches the canonical text instead of paraphrasing from memory.

Bake your judgement into the prep

The AI runs the game the way your prep tells it to. So the rulings you care about belong in the notes, written as instructions next to the scenes they apply to, not left for the moment. This is the part that makes an AI-run table feel like your table.

Each of these costs a sentence, and each is what separates a session that flows from one that stalls on a bad roll.

• Call for a roll only when failure is interesting. Note where a check matters and what is at stake; let routine competence just succeed.
• Fail forward. Write the failure outcome as a cost in time, position, or a resource, never a dead end. Nothing-happens is the worst result a check can have.
• Give plot-critical clues more than one route. Any fact the story depends on needs at least two or three independent ways to surface, so a single missed roll never buries the thread.
• Telegraph irreversible danger in the fiction before it triggers: a trap, an auto-combat threshold, a point of no return. Players may walk in anyway; that is informed agency.
• Let broken enemies break. Unless the module says fight to the death, note that bloodied or leaderless foes flee, surrender, or bargain. Survivors carry hooks.

What the AI handles, and what stays yours

Picture a clean example. The party is one day out from a roadside inn. You prep the inn as a journal page: the innkeeper's name, what he knows about the rival mercenary crew working the north road, and the rumour he shares if asked. You give that innkeeper a written character on his sheet: who he is, what he wants, how he speaks. You pull the rival captain's statblock from a compendium and note she is on deck if the night turns ugly. Your outline says the party arrives tired and short on coin.

At the table, a player asks the innkeeper about the road. The AI reads his page, answers in his words, fills the small edges of the conversation itself, and voices him with a distinct AI voice if you assigned one. Then the rivals draw steel. Now it rolls initiative, plays their turns, tracks conditions, and applies the rulings you wrote down. None of it is invented, because all of it was anchored. Give the AI the anchor and it fills the edges. Characterisation and table-texture are reliable when they sit on top of your prep; long-horizon plot is not, which is why the plot stays on the page.

So you hold the spine and the rulings. The AI holds the bookkeeping, the cast you wrote, and the recall: it references your prep at the table and runs the DM's side from it. It does not author the campaign spine, and it does not run the adventure on your behalf.

Start with one session

You do not need the whole module prepped to begin. Pick a published adventure you own, import it so the journals land in Foundry, pull the first chapter's statblocks from a compendium, and write a one-page outline for the opening session. That is enough to play. Build the next chunk before the next game, not all at once.

Install Familiar in Foundry and connect the AI you already use. If you want a second pair of eyes on structuring a tricky chapter, or you are deciding what goes in the outline versus the journals, the Discord is the place to ask.