AI Roleplay Glossary: 30 Key Terms Every Character Chat User Should Know
A plain-English glossary of AI roleplay and character chat terms, from context windows and lorebooks to OOC, personas, swipes, and character cards.
The entries below are preserved in their original source language to avoid unreviewed machine translation.
This glossary defines the core vocabulary of AI character chat, grouped into model basics, character building blocks, memory, scene play, roles, moderation, and tools. The terms matter because they explain why a character behaves a certain way, why it forgets, and how you steer a roleplay session.
What is a lorebook in AI roleplay?
A lorebook, sometimes called world info, is a set of entries that inject background facts into an AI roleplay session only when their keywords appear. Instead of permanently filling the context window, each entry holds details such as a place, person, or rule and is added on demand. This lets a character recall a large fictional world without spending tokens on facts the current scene does not need.
What is a context window in AI chat?
A context window is the maximum amount of text, measured in tokens, that a language model can consider at once when generating a reply. It includes the system prompt, the character definition, and recent messages. When a conversation grows past this limit, the oldest content is dropped or summarized, which is the main reason an AI character can appear to forget earlier parts of a long roleplay.
What does OOC mean in roleplay?
OOC stands for out of character. It marks text that comes from the user or author as themselves rather than from the persona they are playing, and is used to give directions, ask questions, or set boundaries without disrupting the story. OOC notes are often wrapped in brackets or parentheses. Its opposite is IC, or in character, where everything is spoken as the role.
What is a persona in AI character chat?
A persona is the identity the user adopts inside an AI character chat, describing who they are within the story. It typically includes a name, basic traits, and background, and is supplied to the model so the AI character can address the user consistently and react to them as a defined participant rather than an anonymous voice. A persona shapes the user side of a scene the way a character card shapes the AI side.
What is a character card?
A character card is a structured bundle of data that defines an AI roleplay character. It commonly contains a name, a personality description, a scenario, a greeting, and example dialogue. The card is fed to the language model so the AI can stay in character across a conversation. Cards can be shared as files, letting one author's character be imported and reused by other people on compatible platforms.
Key takeaways
- Most AI roleplay vocabulary comes from three places: machine learning, tabletop and online roleplay, and character-card sharing communities.
- Context window and memory are the terms that best explain why an AI character forgets, so they matter most to new users.
- Character cards and their building blocks, including greeting, scenario, personality, and example dialogue, define how an AI behaves.
- IC and OOC let you separate story dialogue from real instructions, which is the simplest way to steer a scene.
- On story-first apps like OnlyKin, persona, lorebook, and memory work together so a long roleplay stays coherent over many sessions.
Model and generation basics
LLM (large language model): A large language model is an AI system trained on vast amounts of text to predict and generate human-like language one token at a time. It is the engine behind AI character chat, producing each reply by estimating the most likely continuation of the conversation given everything it has been shown so far.
Token: A token is a small unit of text, often a word fragment or a handful of characters, that a language model reads and writes one piece at a time. Token counts measure how much text fits in a context window and how long a response is, so prompts and replies are limited and often billed in tokens rather than words.
Context window: A context window is the maximum amount of text, measured in tokens, that a model can consider at once when generating a reply. It holds the system prompt, the character definition, and recent messages. When a conversation exceeds this limit, older content is dropped or summarized, which is why a character can seem to forget earlier events.
Temperature: Temperature is a setting that controls how random or predictable a model's output is. Lower values make replies more focused and repetitive, while higher values make them more varied and surprising. In roleplay, a moderate temperature is often chosen to keep a character creative without becoming incoherent or drifting off topic.
System prompt: A system prompt is a set of instructions given to the model before the conversation begins, shaping its overall behavior, tone, and rules. In character chat it often defines that the AI should stay in character and follow the supplied persona and scenario. Users usually do not see it, but it strongly influences every reply.
Character building blocks
Character card: A character card is a structured bundle of data defining an AI roleplay character, typically including a name, personality, scenario, greeting, and example dialogue. It is fed to the model so the AI stays consistent across a conversation, and it can be shared as a file so other people can import and reuse the same character.
Greeting (opening message): A greeting, also called the opening message or first message, is the character's initial line that starts a chat before the user types anything. It sets the scene, voice, and tone, establishing where the story begins and what role the user is expected to play.
Scenario: A scenario is the situational context for a roleplay, describing the setting, circumstances, and starting premise of the scene. It tells both the model and the user where and when the story takes place and what is happening as it opens, giving the conversation a clear frame to build on.
Personality: Personality is the part of a character definition that describes traits, temperament, speech patterns, and motivations. It guides how the AI reacts, what it values, and how it speaks, helping the model keep a consistent voice rather than drifting between turns. It is usually written as a description or a list of attributes.
Example dialogue: Example dialogue is a set of sample exchanges included in a character definition to demonstrate how the character speaks and responds. By showing the model concrete instances of the desired style and tone, it anchors voice and behavior more reliably than description alone, especially for distinctive or unusual characters.
Memory and world state
Memory/summary: Memory in AI chat refers to information about earlier parts of a conversation that is retained and reused so the model can stay consistent. Because the context window is limited, many systems compress older messages into a running summary, preserving key facts and events while freeing space for new dialogue in a long roleplay.
Lorebook (world info): A lorebook, sometimes called world info, is a collection of entries that inject background facts into a session only when their keywords appear. Each entry holds details such as a place, person, or rule and is added on demand, letting a character recall a large fictional world without permanently consuming tokens on facts the current scene does not need.
Persistence: Persistence is the degree to which a character's state, memories, and relationship details carry over between sessions rather than resetting each time. Higher persistence means a long-running roleplay can continue where it left off, with the character recalling past events, while low persistence treats each new chat as a fresh start.
RAG (retrieval-augmented generation): Retrieval-augmented generation is a technique where relevant information is fetched from an external store and added to the prompt before the model replies. In roleplay it can surface past events or world facts that no longer fit in the context window, grounding responses in stored detail rather than relying on the model's memory alone.
Playing the scene
IC (in character): IC stands for in character. It describes text spoken or written as the role being played, where dialogue, actions, and reactions all belong to the fictional persona rather than the real person behind it. Staying IC means everything contributed fits the character and the story rather than commenting on it from outside.
OOC (out of character): OOC stands for out of character. It marks text from the user or author as themselves, used to give directions, ask questions, or set boundaries without disrupting the story. OOC notes are often wrapped in brackets or parentheses so they are clearly separate from in-character dialogue.
Swipe (regenerate): A swipe is a request for the model to regenerate its previous reply, producing an alternative version of the same turn. The name comes from mobile interfaces where you swipe between options and keep the one you prefer. Swiping changes only the AI output, leaving the user's own message unchanged.
Narration vs dialogue: Narration is text describing actions, settings, and events, while dialogue is the words characters actually speak. Roleplay typically mixes the two, often marking narration with asterisks or plain prose and placing spoken lines in quotation marks. The distinction helps both the model and the user tell description apart from speech.
Roles and identity
Persona (user persona): A persona is the identity the user adopts inside a chat, describing who they are within the story through a name, traits, and background. It is supplied to the model so the AI can address and react to the user consistently as a defined participant rather than an anonymous voice, shaping the human side of a scene.
Creator: A creator is a person who designs and publishes AI characters, writing their personality, scenario, greeting, and example dialogue. On platforms that allow sharing, creators make character cards available for others to discover and chat with, and a single creator may maintain many distinct characters across different settings and genres.
Character vs companion: A character is any AI role defined for roleplay, often fictional and built around a specific scenario or story. A companion usually describes an AI designed for ongoing personal interaction, emphasizing a continuing relationship rather than a particular plot. The terms overlap, and a given app may lean toward story-driven characters, persistent companions, or both.
Content and moderation
Filter: A filter is an automated system that detects and blocks content a platform does not permit, such as certain explicit or harmful material. It can act on user input, model output, or both, and its strictness varies by service. Filters are a core part of how platforms enforce their content policies.
NSFW: NSFW means not safe for work and refers to content that is sexual, graphic, or otherwise unsuitable for public or professional settings. In AI roleplay it usually denotes mature or explicit material. The label describes content type only and says nothing about whether a given platform allows it.
SFW: SFW means safe for work and refers to content with no explicit sexual or graphic material, suitable for general audiences and public settings. It is the counterpart to NSFW and describes the content type rather than implying any judgment about other material.
Jailbreak: A jailbreak is a prompt designed to make a model ignore its safety rules or restrictions and produce output it would normally refuse. In roleplay the term usually refers to attempts to bypass a content filter. Whether jailbreaking is allowed depends entirely on a platform's policies and terms of service.
Age-gating: Age-gating is a control that restricts access to content or an entire service based on the user's stated or verified age. It ranges from a simple confirmation prompt to stronger verification, and it is used to keep mature material away from minors and to comply with applicable rules.
Formats and tools
Character card PNG/JSON: Character cards are commonly stored as JSON data or as PNG images with the character definition embedded in the file's metadata. The PNG format lets a single image double as both an avatar and a portable definition, so importing the picture also imports the character's personality, scenario, and example dialogue on compatible apps.
SillyTavern: SillyTavern is a popular open-source front-end for AI roleplay that connects to various language models and supports features such as character cards, lorebooks, and personas. It is widely used in hobbyist communities and has influenced common conventions and file formats, which is why many of the terms in this glossary appear across other platforms too.
Multimodal: Multimodal describes an AI system that can work with more than one type of input or output, such as text combined with images. In character chat, multimodal features might let a character interpret an uploaded picture or generate an image alongside text, extending roleplay beyond words alone.
Group chat: A group chat is a roleplay session that includes more than one AI character at the same time, so several defined personalities respond within a single conversation. It allows multi-character scenes where the user interacts with a small cast, and the system manages whose turn it is to reply.
FAQ
What does swipe mean in AI roleplay?
Swipe means asking the model to regenerate its last reply to produce a different version of the same turn. The name comes from mobile interfaces where you swipe sideways between alternative responses, then keep the one you prefer. It does not change your own message, only the AI output.
What is the difference between a persona and a character?
A persona is who the user plays in the story, while a character is the AI-controlled role defined by a character card. The persona governs the human side of the conversation, and the character governs the AI side. Both are descriptions fed to the model so each participant stays consistent.
What is a token in AI chat?
A token is a small chunk of text, often a word fragment or a few characters, that a language model reads and writes one unit at a time. Token counts measure how much fits in a context window and how long a reply is, so both the prompt and the output are billed and limited in tokens rather than words.
What is the difference between SFW and NSFW?
SFW means safe for work and refers to content with no explicit sexual or graphic material. NSFW means not safe for work and refers to content that is sexual, graphic, or otherwise unsuitable for public or professional settings. The labels describe content type only and carry no judgment about whether such content is allowed on a given platform.
What is a greeting in a character card?
A greeting, also called the opening message or first message, is the character's initial line that begins a chat. It sets the scene, voice, and tone before the user has typed anything, and a strong greeting establishes where the story starts and what role the user is expected to take.
What is a jailbreak in AI roleplay?
A jailbreak is a prompt crafted to make a language model ignore its safety rules or content restrictions and produce output it would normally refuse. In roleplay contexts the term usually describes attempts to bypass a content filter. Whether jailbreaking is permitted depends entirely on the platform's policies and terms of service.