Managing Context

For a deeper dive into how context works in Cursor, check out our Working with Context guide.

​

Menu

Using the context menu you can select and pick what you want to send to the model.

​

States

​

Condensed

When files or folders are too large to fit within the context window, Cursor automatically condenses them. Condensing shows the model key structural elements like function signatures, classes, and methods. From this condensed view, the model can choose to expand specific files if needed. This smart condensing approach helps maximize the effective use of the available context window.

​

Significantly Condensed

When a file name appears with a “Significantly Condensed” label, it means the file was too large to include in full, even in condensed form. Only the file name will be shown to the model.

​

Not included

When a warning icon appears next to a file or folder, it indicates that the item is too large to be included in the context window, even in condensed form. This helps you understand which parts of your codebase are currently accessible to the model.

​

Summarization

All models have a finite context window (e.g. 200k tokens), and to provide an answer, the model has to fit all input (which is your user messages, the attached files and images, etc.) and the output (the answer and any tool calls) into this window. If a conversation goes on for too many messages, the context window will fill up until there is literally no space left for the model to respond.

When the context window is about to be filled up, Cursor will automatically summarize the conversation to make sure the model has enough space to respond. We try to keep as much important information as possible, but there are necessarily some things that we have to leave out. You likely will see the model seemingly forgetting some information from earlier, re-reading the same information again, or similar. This is not ideal, but better than not being able to respond at all.

​

What can I do to avoid summarization?

There are quite a few things you can try:

• Start a new conversation if you are working on a new task. You will generally get much better results if there is no unrelated information from earlier tasks in the context, and you are much less likely to hit the context window limit.
• Use a model with a larger context window (see the Models page for a list of models and their context window sizes).
• Include less explicit context (e.g. attach fewer files), and let the agent decide if it needs to read more files.
• Switch to MAX mode, which generally has a larger context window.