Model behavior

Context window

A context window is the maximum amount of text a language model can consider at once when generating a reply.

Обновлено 2026-06-04Копия в Markdown

Краткий ответ

In AI roleplay, the context window contains the system prompt, character card, persona, summaries, and recent messages. When it fills up, older details must be dropped, summarized, or retrieved another way.

Простыми словами

The model does not remember everything by default. It can only respond from the information placed in front of it for the next turn.

Most complaints about characters forgetting names, promises, locations, or earlier scenes trace back to context-window limits and weak memory design.

OnlyKin should explain context windows plainly so users understand why compact memory and structured cards matter.

Источники

OpenAI Help: What are tokens?Used for explaining tokens as the units models process and why input length matters.OnlyKin guide to why AI characters forgetUsed for roleplay-specific context-window examples.