SillyTavern Lorebook Guide (2026): World Info & Triggers

People searching SillyTavern Lorebook or World Info usually need a behavior answer, not a glossary. They want to know why an entry fires, how many recent messages are scanned, where lore is inserted and why a huge book makes replies worse. This guide focuses on that operating loop. It complements the Janitor AI Lorebook guide, the Janitor AI export guide and the AI character card guide without repeating their platform or file-migration focus.

What a SillyTavern Lorebook actually does

SillyTavern uses the names Lorebook, World Info and World Info file for the same general mechanism: a collection of entries that can be inserted into a prompt when matching keys appear in recent context. The entry content is what the model reads; the key and editor comment are activation controls, not automatically visible explanations. World Info guides a model but does not guarantee that the model will mention every fact. Keep identity, current scene state and optional world facts in separate layers so one tool does not become an untestable block of prose.

Design keys that activate predictably

A reliable key is distinctive enough to identify the intended concept but narrow enough not to match ordinary conversation. Prefer a proper place, faction or full name over a common word such as “home.” Add a second key only when it represents a real synonym. Secondary keys should express a condition, not become a second unrelated topic. Regex can be useful for controlled patterns, but it increases the test surface. Every key should have a simple message that activates it and another message that should not.

The settings that change activation

Scan depth, token budget, case sensitivity, whole-word matching, insertion position, order, probability and recursion all change behavior. Start with default insertion and a small budget. During debugging, scan only the last few turns so you can explain why an entry fired. Whole-word matching is useful for English single words but can be awkward for languages without whitespace. Constant entries should be rare because they spend context in every request. A larger budget is not a substitute for concise content.

Add one entry before building a large book

Build a tiny book first. Write one standalone entry that names the subject, its important facts and the conditions under which it matters. Do not rely on the editor label or key to provide context. Save a JSON backup before editing, and keep a changelog with entry count, key changes and settings. After one entry passes, add a second entry with a different key. Bulk imports hide failures because a missing key, wrong scope or oversized content can affect dozens of entries at once.

1. Create or open the World Info file.
2. Add one concise entry with one distinctive key.
3. Save, open a disposable chat and mention the key once.
4. Inspect the generated prompt or activation indicator.
5. Add a second entry only after the first behaves correctly.
6. Keep a JSON backup before bulk edits.

Choose Character, Chat, Persona or Global scope

Use Character Lore when facts should follow one character across that character’s chats. Use Chat Lore for a single branch, campaign or temporary state. Use Persona Lore for information attached to the user persona. Use Global World Info only when cross-character reuse is intentional. The narrowest scope makes debugging easier and reduces accidental context leakage. When exporting a character, keep a standalone backup because a bound book and an embedded card are not always the same artifact.

Run an eight-turn World Info test

Run the same eight-turn test after creating, importing or changing a book. Test a message with no keys before testing exact activation. Then vary case, partial words, two simultaneous keys, secondary conditions, scene changes and a reopened chat. Record whether the entry activated, where it appeared, how many tokens it used and whether the reply stayed coherent. If you change model, preset, extension or scan settings, treat that as a new test rather than comparing memories of an earlier run.

Fix missing, noisy or misplaced lore

Debug from activation to generation. If an entry never fires, check spelling, enabled state, binding and scan depth. If it fires too often, narrow the key, turn off constant or inspect recursion. If it activates but the model ignores it, shorten content and make it standalone. If imported lore is truncated, split it by concept and lower repetition. The Janitor AI Lorebook export guide covers authorized conversion; if your goal is immediate character chat without maintaining World Info files, try the LumiChat character library.

FAQ

Is a SillyTavern Lorebook the same as World Info?

Yes. World Info is SillyTavern’s current term for the dynamic lorebook system.

How do Lorebook triggers work?

A key matches the scanned chat context and inserts that entry’s content into the prompt.

What scan depth should I use?

Start with two to four recent turns while debugging, then increase only when needed.

Why does my entry activate too often?

The key may be broad, constant may be enabled or recursive activation may be pulling more entries.

Should I use Character Lore or Chat Lore?

Use Character Lore for durable character facts and Chat Lore for one branch or temporary state.

Does World Info guarantee the model will remember a fact?

No. It adds context; the model can still omit, misunderstand or contradict it.

Can I use regex keys?

Yes, when you understand JavaScript regex behavior and test false positives.

Can LumiChat import my SillyTavern Lorebook?

No direct import is promised here. LumiChat is a separate hosted character-chat service.