Glossary¶

Key terms used throughout Lumiverse and these guides.

A¶

Alternate Fields: Variant versions of a character's description, personality, or scenario that can be selected per-chat.
Alternate Greetings: Multiple first-message options for a character. Chosen when starting a new chat.
API Key: A secret credential from an AI provider that authenticates your requests. Stored encrypted in Lumiverse.
Assembly: The process of building the complete prompt from preset blocks, character data, world info, macros, and chat history.
Author's Note: A hidden instruction injected at a specific depth in the conversation to influence the AI's behavior.

B¶

Block (Prompt Block): A section of text in a preset that contributes to the assembled prompt. Blocks have roles, positions, and can be enabled/disabled.
Branch: A fork of a conversation at a specific message, creating a new chat with shared history up to that point.

C¶

Character Card: A standardized format (CCSv1/v2/v3) for packaging character data, optionally embedded in a PNG image.
CHARX: A ZIP archive format for character cards that can include the card JSON, avatar, expressions, and Lumiverse modules.
Connection: A saved configuration linking Lumiverse to an AI provider (includes provider, model, API key, and URL).
Constant Entry: A world book entry that is always included in the prompt regardless of keyword matching.
Context Window: The maximum number of tokens a model can process at once (prompt + response combined).
Cortex (Memory Cortex): Lumiverse's narrative-aware memory layer that tracks entities, relationships, salience, and arcs on top of basic long-term memory. See Memory Cortex.
Council: A multi-persona deliberation system where AI personas analyze the scene and provide guidance before the main generation.

D¶

Databank: A knowledge bank of uploaded documents that can be RAG-retrieved during generation. Documents are referenced with #slug mentions in chat. See Databank.
Depth: How many messages from the end of the chat to insert content. Lower depth = closer to the end = more influence.
Dry Run: A test that assembles the full prompt without actually calling the AI. Shows exactly what the model would see.

E¶

Embedding: A numeric vector that represents the meaning of a piece of text. Powers semantic world-book activation, long-term memory, databank retrieval, and the cortex. See Embeddings.
Entity: A named thing tracked by the Memory Cortex — character, location, faction, item, concept, or event. Accumulates facts, relationships, and salience over time.
Expression: An emotion-mapped image that changes dynamically based on the conversation mood.

G¶

Generation: The process of sending a prompt to the AI and receiving a response. Types: normal, regenerate, continue, swipe, impersonate, quiet.
Global Add-On: A persona add-on stored in a shared library, attachable to any number of personas. Edits propagate everywhere it's attached.
Group Chat: A conversation with multiple AI characters who take turns responding and interact with each other.

I¶

Interlink: A live, bidirectional link between two chats' Memory Cortex state. Each chat sees the other's entities and relationships in real time.

L¶

Loom: Content blocks from packs, categorized as narrative styles, utilities, or retrofits.
Lorebook: Another name for World Book — a collection of keyword-triggered contextual entries.
Lumia: An AI persona from a pack, used as a council member or for narrative style selection.

M¶

Macro: A template variable (e.g., {{char}}) that gets replaced with dynamic content during prompt assembly. See Macros Reference.

O¶

Operator Panel: Settings panel for instance-level operations — check for updates, switch git branches, restart the server. Requires the runner to be attached.
Outlet: A named content slot a world book entry can export — referenced from presets and other entries via {{outlet::name}}.

P¶

Pack: A content bundle containing Lumia items, Loom items, and/or Loom tools.
Persona: Your identity in conversations — includes name, pronouns, description, avatar, and optional add-ons.
Persona Add-On: An optional, toggleable block of content attached to a persona. Lets you extend your persona description dynamically without editing it.
Preset: A saved configuration defining prompt block order, sampler settings, and completion behavior.
Preset Profile: A snapshot of block enabled/disabled states, bindable to default, character, or chat contexts.
Provider: An AI service (OpenAI, Anthropic, Google, etc.) that Lumiverse connects to for generation.

R¶

Recursion (World Info): When activated world book entries contain keywords that trigger additional entries.
Regex Script: A text transformation rule using regular expressions, applied at various stages of the pipeline.
Runner: The supervisor process that the start scripts attach. Performs updates, branch switches, and restarts on behalf of the Operator Panel. Disable with --no-runner if you don't want it.

S¶

Salience: A score (0.0–1.0) the Memory Cortex assigns to each memory chunk based on emotional weight, narrative flags, and information density. High-salience memories resist decay.
Sampler Settings: Parameters that control how the AI generates text (temperature, top-p, penalties, etc.).
Scan Depth: How many recent messages are checked for world book keywords.
Selective Logic: Secondary keyword conditions (AND, OR, NOT) that refine when a world book entry activates.
Sidecar: A secondary, usually lighter LLM used for background tasks — Memory Cortex extraction, council tools, expression detection, heuristic arbitration.
Sovereign Hand: A Loom co-pilot mode that lets you steer the next generation with directorial instructions while the AI keeps narrating.
Spindle: Lumiverse's extension system. Extensions run in an isolated Bun Worker with a permission-gated RPC bridge. See Extensions.
Spindle Capability: A manifest declaration that bypasses install-time code-pattern scans for legitimate uses of patterns like eval or base64 decode. Distinct from runtime permissions.
Spindle Permission: A runtime grant that lets an extension access a specific subsystem (chats, generation, cors_proxy, etc.). Privileged permissions require admin approval.
Sticky Entry: A world book entry that stays active for a set number of turns after its keywords stop appearing.
Swipe: An alternate version of a message. Generated by regenerating a response, creating a deck of alternatives.

T¶

TagLibrary: A SillyTavern extension that stores character tags separately from the card. Lumiverse's migration wizard can re-apply TagLibrary backups after a regular SillyTavern import.
Theme Pack: A shareable bundle containing theme variables, custom CSS, component overrides, and uploaded theme assets. Exported and imported from the Custom CSS modal.
Token: The basic unit of text for AI models. Roughly 3/4 of a word. Used to measure context size and response length.

V¶

Vault: A frozen snapshot of a chat's Memory Cortex state (chunks, entities, relationships) packaged as a reusable, read-only knowledge object. Attach it to other chats so they "remember" the source chat.

W¶

World Book: A collection of entries that inject contextual information into the prompt when their keywords appear in the conversation. Also called a lorebook.
World Info: Another name for World Book content and the system that activates it.