How to Manage Claude Tokens
◆ Token Basics

Context window explained

What "context window" means

The context window is the maximum number of tokens Claude can "see" at once — both your input and its output combined. Claude Sonnet supports 200K tokens. Claude cannot reference anything outside this window.

What fills the window

  • Your system prompt (runs every turn)
  • The full conversation history
  • Any documents or files you paste in
  • Tool definitions and results
  • Claude's response

Once you hit the limit, older messages get truncated or the API returns an error. Long multi-turn conversations are the #1 silent credit drain in production apps.

operator note

Build a conversation summarization step that fires before you hit 50% of the context window. This keeps costs predictable.

Changelog · 1
  • Initial release — 5 sections, 11 lessons.