Claude Code /compact vs /clear: When to Use Each
Use /clear when you switch to unrelated work, because it starts a new conversation with empty context.
Use /compact when you want to keep working on the same task, because it replaces the conversation with a summary to free up space.
Both commands reduce what Claude has to read on every message, which saves tokens and helps Claude focus. They just do it in different ways.
This article explains what each command does, when to use it, and what happens to your files, memory, and instructions.
/clear vs /compact at a glance
/clear | /compact | |
|---|---|---|
| What it does | Starts a new, empty conversation | Summarizes the current conversation |
| Claude remembers | Nothing from the old conversation | A summary of it |
| Best for | Switching to unrelated work | Continuing a long task |
| Old conversation | Saved, and you can resume it | Replaced by the summary |
When to use /clear
Run /clear when the next task has nothing to do with the last one:
/clear
Old conversation takes up space and costs tokens on every message, and it can distract Claude from the files you need next.
Your old conversation isn’t deleted. You can bring it back with /resume. To make it easier to find, give it a name as you clear:
/clear release-prep
/clear also has two aliases: /reset and /new.
When to use /compact
Run /compact when you’re deep into a task and want to keep going:
/compact
You can tell Claude what to keep in the summary:
/compact focus on the auth bug fix
This works better than letting Claude guess what’s important, so it’s a good idea to run it before starting a long new step.
Claude Code compacts automatically
You don’t have to wait for problems. When your conversation gets close to the context limit, Claude Code compacts it automatically, so a full context window doesn’t end your session.
To compact earlier, set how full the context can get before it runs:
/autocompact 500k
To see how much of your context is being used, run /context.
What survives compaction?
After compaction, Claude Code reloads some things from disk, so they’re not lost in the summary:
| Content | After compaction |
|---|---|
Project-root CLAUDE.md | Reloaded from disk |
| Auto memory | Reloaded from disk |
| Your plan from plan mode | Reloaded from disk |
| Files Claude read or edited | Up to five of the most recent are re-read |
| Skills Claude used | Reloaded, but large skills are cut short |
| Everything else in the conversation | Summarized |
Since CLAUDE.md is reloaded, put instructions you always want Claude to follow there. See where Claude Code memory is stored for how to set it up.
Compact only part of the conversation
You can also summarize just part of the conversation. Run /rewind, choose a message, then pick one of these:
- Summarize from here: summarizes everything after that message
- Summarize up to here: summarizes everything before that message and keeps later messages as they are
How to get back a cleared conversation
Run /resume and choose the old conversation from the list. If you named it with /clear <name>, search for that name.
In the same Claude Code session, you can also open the rewind menu with /rewind and pick the previous conversation from there.
Conclusion
Use /clear between unrelated tasks and /compact to keep going on a long one. Either way, your instructions in CLAUDE.md stay in place.
Keeping your context small also makes Claude faster and uses less of your plan. See why Claude Code is so slow for more tips.