Context Window

🎥 Video lesson — video not yet migrated (source Skool video id: af169eb12ada4d4cb13da781b7b15de7).

Understanding how Claude Code's memory works will make you dramatically better at using it. Claude Code has what's called a context window — its working memory during one conversation. Every message you send, every response it gives, every file it reads, every plan it creates — all of that goes into the context window. Every Claude model currently has a 200,000 token context window (roughly 150,000 words). Sounds huge, but on a big project it fills up faster than you'd expect. You can check your usage anytime by typing /context.

When the context fills up to about 80%, Claude Code triggers "compaction" — it compresses the conversation into a shorter summary and frees up space. Like taking meeting notes and clearing the whiteboard. But compaction isn't lossless. Details get lost. And each time it happens, output quality degrades a little more.

This is exactly why CLAUDE.md is so powerful — it's permanent context that never gets compacted. It's also why advanced users use agents: instead of one long conversation that fills up and gets compressed, you have multiple focused conversations that each start fresh with a clean context window. The number one rule: don't try to do everything in one conversation. Building a website and then creating email campaigns in the same session will hurt your results. Break work into separate sessions for dramatically better output.