Lorebooks (World Info)

A lorebook is a named collection of trigger-keyword entries that Pyre injects into the prompt only when they become relevant — so the model recalls persistent world facts without you spending context on them every turn.

What a lorebook is

Think of a lorebook as a reference binder for a setting: a list of small entries, each one a fact, place, person, or rule. Each entry watches for its own keywords. When one of those keywords shows up in the recent conversation, that entry's text is slipped into the prompt for the next turn. The rest of the time it stays out of the way, leaving room in the context window for the actual scene.

This is why lorebooks scale: a world can hold dozens of entries, but only the handful that matter right now are ever sent.

Entry fields

Each entry in a lorebook has these fields:

Field	What it does
Keys	The trigger keywords. When one appears in the recent message window, the entry fires.
Content	The text injected into the prompt when the entry fires.
Constant	An always-on flag. A `constant` entry fires every turn, regardless of keywords — for facts that must always be in context.
Enable / disable	Turn an entry on or off without deleting it. Disabled entries never fire.
Order (priority)	Sort order for fired entries. Hits are sorted by priority so the most important lore lands in a predictable place.

How keyword scanning works

Before each turn, Pyre scans the recent message window (by default, the last 6 messages) for entry keywords. The scan is case-insensitive, so Vael and vael match the same key.

Entries whose keywords appear in that window fire for the turn.
constant entries always fire, whether or not their keywords showed up.
The fired entries are sorted by priority (the order field) before injection.

Because the scan looks only at recent messages, lore surfaces when it comes up in conversation and naturally drops out again when the topic moves on.

Where the active books come from

Pyre pulls lore from three binding sources, combined additively and deduped into one set for the chat:

Per-chat attached — books you attach to this specific chat.
Character-bound — books bound to a character, automatically active whenever that character is in the chat.
Persona-bound — books bound to your persona, active whenever that persona is in play.

In group chats, every member's bound books participate, so each character brings its own world facts to the scene regardless of who is currently responding.

Info

Worlds travel with the characters and personas that own them. Bind a lorebook once to a character, and it follows that character into every chat — no need to re-attach it each time.

Per-chat override

You can disable an inherited book for one chat — a character- or persona-bound book — without unbinding it globally. The book stays bound everywhere else; it just sits out this one conversation. Explicit per-chat attachments are unaffected by this override.

Embedded books: import and export

Cards can carry their own lorebook in the standard character_book field.

On import, you can keep an imported card's book embedded only — bound to that character and active, but hidden from the main management list (still re-attachable later).
On export, all of a character's bound entries merge back into the PNG's character_book, so other apps in the ecosystem see the world too.

This gives a full round-trip: a card you import with a world keeps that world, and a card you export carries its world to wherever it goes next.

Diagnostics

A scan trace shows which entries influenced a turn — for example "book • matched keyword" or "constant entry", along with how many entries were scanned, how many fired, and how many were skipped because they were disabled. Use it to debug why a piece of lore did or didn't show up.

Managing lorebooks

Lorebooks are importable and exportable on their own, and you manage them under More → Lorebooks.