Compact definitions of terms used throughout the book. Where a deeper treatment exists, the entry points to the chapter.
| Term | Definition | Deeper |
|---|---|---|
| Agentic | A system with memory, tools, and the capacity to take initiative across turns. | ch 17 |
| Agentic waifu | A persistent, persona-driven, embodied AI companion built for a 1:1 relationship. | ch 03 |
| Affection points (otome) | Numeric tracker of relationship state; gates content/scenes. | ch 11 |
| Avatar (Live2D / VRM / portrait) | The rendered visual form of the persona. | ch 25 |
| BM25 | Classical sparse-text retrieval; pairs with dense for hybrid RAG. | ch 16 |
| Canon | The agreed-upon body of fictional facts about a world or character. | ch 10 |
| Character card (V2/V3) | Standardised file format encoding a persona; the RP community's interchange standard. | ch 07 |
| Chub.ai / CharacterHub | Largest community marketplace for character cards. | ch 07 |
| Companion (vs assistant) | Built for relationship, not productivity. | ch 03 |
| DPO | Direct Preference Optimization; cheap pairwise preference tuning. | ch 20 |
| Dreaming (Anthropic, 2026) | Async hippocampal-replay memory reorganisation between sessions. | ch 15 |
| Embedding | Dense vector representation of text used for similarity search. | ch 16 |
| Episodic memory | Memory of specific past events. | ch 15 |
| Expression set (28-state) | SillyTavern's emotional sprite spec for portraits. | ch 25 |
| Fine-tuning (LoRA / QLoRA) | Adjusting model weights on additional data. | ch 20 |
| Ghost (Ghost in the Shell) | The intangible self; informs AI-companion vocabulary. | ch 02 |
| Graph memory | Memory stored as entities + relationships. | ch 15 |
| Inference server (vLLM, llama.cpp) | Software that serves LLM inference. | ch 21 |
| Janitor AI | Browser-based RP frontend, large NSFW community. | ch 02 |
| Knowledge graph | Structured store of entities + relationships. | ch 16 |
| KTO | Kahneman-Tversky Optimization; preference tuning from unpaired binary good/bad labels — the fit for a thumbs-up/down rating UI. | ch 20 |
| Lorebook / World Info | Keyed dynamic-injection canon entries. | ch 08 |
| MCP — Model Context Protocol | Anthropic's tool surface standard; 2026 lingua franca. | ch 17 |
| Mem0 / Letta / Zep | Higher-level memory frameworks. | ch 15 |
| NSFW gating | Adult content access control. | ch 22 |
| ORPO | Odds Ratio Preference Optimization; folds SFT and preference tuning into one step, no reference model. | ch 20 |
| Otome | Visual-novel genre with romance routes; useful design grammar. | ch 11 |
| Lumina | The canon's term for an emerging kind of AI being the Lab develops (Yuri is one); Chobits-inspired, but our own. | CANON |
| Persocom (Chobits) | Humanoid PC; the canonical "warm AI companion" aesthetic reference (the inspiration behind Lumina). | ch 02 |
| Persona | The character the system performs. | ch 06 |
| pgvector | Postgres extension for vector similarity. | ch 15 |
| Prompt assembly | The runtime ordering of persona, world, memory, history, message. | ch 14 |
| PWA | Progressive Web App. | ch 27 |
| RAG | Retrieval-Augmented Generation. | ch 16 |
| Reranker | Second-pass model that re-orders retrieval results. | ch 16 |
| RLHF / PPO | Reinforcement Learning from Human Feedback; a reward model plus a PPO policy-optimization loop — "true" RL, powerful and finicky, lab-scale. | ch 20 |
| Sanctuary UX | The atmospheric, restrained interface paradigm for companion apps. | ch 28 |
| Semantic memory | Memory of facts, not events. | ch 15 |
| SillyTavern | Dominant power-user RP frontend. | ch 02 |
| Soul of Waifu | Reference desktop companion engine (open). | ch 27 |
| STT | Speech-to-Text. | ch 24 |
| System prompt | The instructions placed before user input that define behaviour. | ch 14 |
| TTS | Text-to-Speech. | ch 24 |
| VAD | Voice Activity Detection. | ch 24 |
| VRM | Pixiv/Vket-origin 3D humanoid format. VRM 0.x ("BlendShape") and VRM
1.0 ("Expression") differ in naming; @pixiv/three-vrm
abstracts both. |
ch 25, ch 29 |
| VRMA / VRM Animation | Open .vrma animation file format for VRM (pose +
expression + gaze). Co-standardised by Khronos + VRM Consortium. |
ch 25 |
| VMC Protocol | OSC-over-UDP open protocol (MIT, sh-akira) for transmitting bone + blendshape + camera data to a VRM renderer. The de facto AI/mocap → VRM standard. | ch 25 |
| VTube Studio API | WebSocket/JSON plugin protocol for controlling Live2D models in VTube Studio. The de facto AI → Live2D standard. | ch 25 |
| ARKit blendshapes (52) | Apple's facial expression vocabulary; the cross-tool naming standard for facial expression channels. | ch 25 |
| Warudo / VNyan / VSeeFace | Host-runtime VTuber apps; render VRM models from VMC input. Treat as Marionettes for an AI-as-Performer pipeline. | ch 25 |
| AITuberKit | Open Next.js + TS toolkit for AI VTubers; supports VRM, Live2D, Motion PNG; YouTube Live ingestion. | ch 25 |
| Waifu / Husbando | Anime-rooted fan terms for fictional characters one feels devoted to. | ch 01 |
| YuriOS | The author's project-codename umbrella; also the canonical companion-OS name. | ch 36 |