Manage context for AI roleplay and long conversations
Budget conversation history, character cards, and response length within a model context window.
Context includes the whole request
The model sees system instructions, character descriptions, lore, prior messages, and the new message. Your visible latest message is only one part of the input. The context cap includes input plus generated output.
Leave room for the response
For a 61,440-token context limit, a 53,248-token prompt leaves at most 8,192 tokens for output, subject to the separate output cap. Client formatting can add tokens, so leave some headroom.
Keep useful history
Trim repetitive turns or summarize older events while keeping current character and scene details. A larger context limit alone does not guarantee better memory or more consistent writing. Check model availability and limits before choosing a client preset.
Faster replies in long chats
Between your messages the model keeps the start of the chat in temporary memory, so the next reply only reads what is new, as long as the start of the prompt stays the same. Anything near the top that changes with every message makes it read the whole chat again, which can add several seconds before the first word. Keep macros such as {{time}}, {{date}}, {{random}} and {{idle_duration}} out of the character card and system prompt, and place lorebook or World Info entries that switch on and off later in the prompt (for example at a chat depth) rather than before the character definition. Once a chat is longer than the context size, the app drops the oldest messages every turn and the whole history is read again each time; summarizing older events keeps replies quick.
Next step
Get your values on Setup, check the models, or make a key in your account.