What Is a Lorebook? The Setting Dictionary Behind AI Character Chat
It feels like a character should get better the more setting you write into it. In practice the opposite happens surprisingly often. The lorebook exists to solve exactly that: instead of carrying every fact into every turn, the character looks up only what the current topic touches.
The one-line definition
A lorebook is a collection of entries. Each entry has a title, a short body, and a list of keywords. When one of those keywords shows up in the recent conversation, the entry body is slipped into what the model reads for that turn. When it does not show up, the entry stays out.
That is the whole mechanism. Everything else is about using it well.
Why it exists
A model reads a limited amount at once. If a character sheet carries three thousand characters of world history into every single turn, that history is competing with the last things you actually said — and the recent conversation is what loses. This is one of the quiet reasons long roleplays start forgetting details.
A lorebook flips the default. The bulk of the setting sits outside the conversation and only the relevant pieces come in. A city gets described when someone mentions the city. The old betrayal comes up when the old friend’s name does.
How it fires on MusePick
- Each entry lists its keywords. Before every reply, the last few turns of conversation (three by default) are scanned for them.
- Matched entries are inserted, along with any entry marked always active — those go in every turn regardless.
- When too many entries match, priority decides which survive; there is a ceiling on how many entries and how many characters go in per turn.
- One lorebook can be attached to several characters, and one character can use several lorebooks. Shared world, separate people.
What belongs in it — and what does not
Belongs: proper nouns and what they mean — places, organizations, past events. History between people. The rules of the world that only matter when they come up.
Does not belong: the character’s personality and way of speaking. Those are needed every turn, so they belong in the character sheet, not behind a keyword. The same goes for anything that must always be present — that is what always active is for, and it should be rare.
Five habits that make it work
- One fact per entry. An entry that explains three things fires for all three and wastes space on two of them.
- Write keywords the way people actually talk. Include the nickname, the abbreviation, the misspelling you expect. “Headquarters” never fires if everyone says “HQ”.
- Keep bodies to two to four sentences. The entry is a reminder, not a chapter.
- Secrets go in the episode’s secret field, not in the lorebook. That field carries a rule the model follows — act on it, never say it. A lorebook entry has no such guard.
- Three always-active entries at most. Every one of them costs room in every turn forever.
The mistakes that break it
Keywords that are too common. An entry keyed to “home” or “he” fires on almost every message, which means it is not a lorebook entry anymore — it is dead weight in the character sheet with extra steps.
Entries written like prose. A four-paragraph backstory in one entry either gets truncated or crowds out everything else that matched.
The same fact in both places. If the character sheet already says she grew up in Busan, a lorebook entry saying it again gains nothing and spends a slot.
Try building one
On MusePick, lorebooks live in the creator studio alongside characters and worlds. Start with five entries — the places and people your story keeps returning to — and watch which ones actually fire in a test chat before you write more.
Open the studio