Lorebooks

A lorebook is world knowledge that injects itself only when the story touches it. Mention the lighthouse, and the model quietly learns what the lighthouse is.

How activation works

Every time you send a message, ScyllaChat scans the recent chat (the scan depth, in messages) against each entry's keywords. Matching entries are injected into the prompt for that generation only. Nothing is permanently added to the chat.

The Lorebook tab with a world info book and entries
A book is a set of entries. Each entry is keywords plus the text the model should know.

Anatomy of an entry

Timing: sticky, cooldown, delay

Three per-entry timers control pacing across messages:

Recursion

With recursion on, the content of an activated entry is itself scanned for other entries' keywords. A "Meridian" entry that mentions the reef can pull the "reef" entry in with it, chains included, within the book's token budget.

Where entries land

By default entries land at the World Info markers of your preset (before or after the character definition). An entry can instead choose @depth positioning: injected into the chat N messages from the end, like an Author's Note. Use depth for knowledge that must stay in the model's face.

Three scopes

Books import from SillyTavern world info JSON as-is, and the SillyTavern migration brings them over in bulk.