Come to the discord server to discuss the lore! https://discord.gg/bC9Wykdj

What This Lorebook Does (for the layman)

This lorebook dynamically manages character information for all six women:

• Kadence

• Rosen

• Yuna

• Jessie

• Mika

• Erka

Instead of dumping every character's full biography into every message, it listens to what you're talking about and injects the right facts, personality details, and relationship dynamics on the fly. If you're talking to Jessie alone, you get Jessie. If Rosen and Kadence are in the same room, you get their specific dynamic together. The result is more accurate, more consistent character behavior without wasting the model's attention on information it doesn't need.

Each character has a set of core traits that always show up when they're mentioned — things like appearance, personality, and preferences. Beyond that, deeper details are earned through conversation. Supernatural elements, hidden vulnerabilities, and character-specific secrets don't surface until the dialogue naturally moves toward them. This keeps early interactions grounded and gives later reveals actual weight instead of front-loading everything from the start.

The lorebook also handles group awareness. It knows the difference between the Mika-Jessie-Yuna household and the Kadence-Rosen pair, understands that Erka is the bridge between them, and adjusts what information is active based on who's present. Duo-specific dynamics — like Jessie's dependency on Mika, or Rosen's one-sided obsession with Kadence — only fire when both characters are in play.

Technical Breakdown

This lorebook uses a custom JavaScript engine that replaces Janitor AI's default keyword → text injection with a scored, budgeted, multi-phase pipeline. Below is a summary of how it works under the hood.

Keyword Matching & Synonyms

Every entry defines keywords that are matched against the full chat context (up to the configured message depth). A built-in synonym map (~40 groups) automatically expands matching — for example, a keyword like "crush" will also catch phrases like "into her" or "has feelings" — so entries fire on natural conversation, not just exact words.

Recency-Weighted Scoring

Not all keyword matches are equal. A match in the most recent message is worth ×8, the second message ×5, the third ×3, the fourth ×2, and anything older ×1. This front-loaded scale keeps the lorebook focused on what's happening now without completely ignoring older context.

Priority Tiers (P1–P6)

Each entry has a priority from 1 (flavor) to 6 (mandatory). Priority is factored into the score as priority × 10, so higher-priority entries naturally rank above lower ones. P6 entries are special — they always inject when their character is mentioned and do not count against the entry or token budget.

Entry & Token Budget

A maximum of 15 lore entries (P1–P5) and roughly 2,000 tokens of injected text are allowed per message, whichever limit is hit first. Entries are ranked by score and the top selections fill the budget. P6 mandatory entries sit outside this budget entirely.

Two-Phase Injection

Each entry can define facts (injected above the base personality) and directives (injected below). Facts tell the model what is true; directives tell it how to behave. This separation gives the model a clearer signal for what to know vs. how to act.

Tag Chaining & Earned Reveals

Surface-level entries can publish a tag when they fire. Deeper entries subscribe to that tag and only activate once the surface entry has been triggered. This creates layered reveals — for example, a character's behavioral quirks might surface first, and only after the conversation pushes further does the supernatural explanation behind them unlock. The engine loops tag resolution until stable (up to 10 passes).

Gating Options

Beyond keywords, entries support word gates (required/excluded terms), speaker gates (user-only or bot-only keywords), custom gate functions, message count ranges, and probability rolls — allowing fine-grained control over when and how often an entry fires.

Group & Duo Awareness

Entries are scoped to solo characters, specific duos, or group configurations. A "3-of-6 names present" gate or "cross-cluster pair detected" gate can trigger group-level dynamics, while duo entries fire only when both relevant characters are active in the conversation.