Memory and companionship glossary
A context window is the limited amount of information a model can work with for a response. It can include instructions, conversation history and retrieved details. A long chat may exceed that working space even if the app still displays every message. Visible history and active model context are different things.
Definition
A context window is a model’s bounded working space for the information used in a generation. Context is usually measured in tokens rather than words and can include instructions, messages, documents and other supported inputs; output can also share the limit, depending on the model and API. Applications may trim, select or summarize older conversation when assembling the next request.
By the Prelulu editorial team · Updated
Defines working context and explains context management for Claude; individual model limits and accounting vary.
A character’s hometown can remain visible far up your chat history while the application sends only more recent messages to the model. The interface has retained the message, but it may no longer be part of the model’s active context for the next reply.
An app can replace a long exchange with a shorter summary to preserve important context. Summaries can omit details or compress ambiguity. Repeating a crucial scene fact may help when the next response depends on something a summary did not preserve.
A larger window can hold more information, but it does not guarantee accurate recall, better roleplay or lower latency. How an app chooses instructions and retrieves memories also matters. Ask what it carries between sessions as well as how much fits in one request.
FAQ
No. Training shapes the model before your conversation. Context is information supplied or assembled for a particular response. Putting a fact in a chat does not by itself mean the model has been retrained on it.
No. Capacity for one generation and storage across sessions are different. A service needs a separate mechanism to save and retrieve information later.
No. A token can represent part of a word, punctuation or another unit used by the model. Token counts vary with language and content, so a word count does not translate into one universal context size.
Choose a character for conversation, roleplay or a live call. Check the current plans and available credits before calling.
Explore AI charactersUpdated
