The context window is the maximum amount of text, measured in tokens, that a language model can read and consider in one request.
Everything the model knows about your current task has to fit inside the context window: the system prompt, the conversation so far, tool results, retrieved documents and the model's own reply. Windows have grown from a few thousand tokens to hundreds of thousands, but they are still finite, and long agent sessions fill them quickly. Once the window is full, something has to be dropped or summarized before the conversation can continue.
The context window is working memory, not storage. It resets with every new conversation and is shared with every tool result and file the model reads. That is why retrieval matters: instead of loading a whole project into the window, a system searches for the few relevant chunks and loads only those. It is also why memory checkpoints matter. Anything worth keeping past the window's lifetime needs to be written somewhere outside it.
How Central Brain handles context window
semantic_search and search_text, with the source path and page number where available, instead of whole files.