Compact

Compact summarizes a long conversation you've had with an AI, keeps only that summary, and deletes the original conversation history to free up the AI's memory space again. Just as a desk piled high with papers makes it hard to find the important document, an AI's memory can fill up as a conversation grows longer, making it easy to lose track of earlier context. Compact acts like a cleaner, tidying that pile of papers into a single summary sheet and clearing the desk.

The catch is that truly important details can get lost during summarization. So rather than just letting it run automatically, it's better to tell the AI directly what to keep and what to discard when running Compact. For example, if you instruct it to drop the debugging process and keep only the API changes, the AI will summarize according to that standard and won't miss what matters.

This is especially useful for long tasks that take hours. You can clear away the initial setup or trial-and-error process and keep only the core content you need to continue, allowing the work to carry on smoothly.

A Claude Academy course introduced Compact as one of the tools for steering long tasks, explaining, "If debugging is done and you only want to keep the API changes, just tell it to do that." Contrary to a common misunderstanding, Compact isn't simply a feature that deletes conversation—it summarizes and rewrites it, so if you don't specify what to keep, important content can disappear along with the rest.

If a conversation has grown long during a lengthy task, write down what you want to keep before running the Compact command.

Example: /compact Clear out the debugging process so far, and keep only the decision to change the API response format and the remaining to-do list.

Giving instructions this specific gives the AI a clear standard for what to keep in the summary, so the work can continue without losing direction.