Glossary

Model : The AI that generates replies. It can run locally (Ollama) or on a remote OpenAI-compatible service.

Provider : The connection to a model: a URL, a type and, if needed, an API key. You can have several providers and choose one per conversation.

Context : Everything sent to the model for a reply: character prompt, memories, summaries, recent messages and the current instruction. It has a size limit measured in tokens.

Token : A small unit of text the model processes. Context limits are usually expressed in tokens, not words.

Memory : A fact the character should remember across messages. Memories can be proposed automatically or approved manually and can decay over time.

Summary : A compressed version of older messages, kept in the prompt instead of the full text. It helps long conversations stay inside the context limit.

Branch : A parallel conversation that starts from a message and inherits the settings of the original. Both histories remain available.

Rewind : Return the conversation to an earlier message and continue from there.

OOC : Out of character. Text outside the character’s voice, usually between //...//, shown as a separate pill.

PWA : A website installed on the phone’s home screen with its own icon and window. Roleplay Manager uses this for phone access; there is no native app.

Portable package : The ZIP or tarball that contains the app, its Node runtime, the launcher and the data/ folder. It does not need an installer.