Guides

What uses your tokens, how to find the session responsible, and what to change.

• How to check your Claude Code usageFour ways to check Claude Code usage: the /usage command, /context, a status line, and the log files. Plus how to see which session used the most.
• How to check your Codex usageCheck Codex usage with /status and /usage in the CLI, the usage dashboard, or the log files. Plus how to see which Codex session used the most.
• A Claude Code status line that shows context and limitsA ready-to-paste Claude Code status line that shows tokens in context, percent of the window used, and your 5-hour and weekly limits.
• Claude Code notifications: know when it needs youSet up Claude Code notifications with a Notification hook so you know when it needs you. Plus what hooks cannot catch: a check that failed while the agent waited.
• Why is Claude Code so expensive?Claude Code re-reads the whole conversation on every turn. Here is what actually uses your plan: long history, overhead loaded before any work, and subagents.
• Which Claude Code session is using my limit?Claude Code shows how much of your limit is used, not which session used it. Here is how to find the session that is eating your tokens.
• Claude Code compact and auto-compact: when to use themWhat /compact and auto-compact do in Claude Code, and when to compact: judge by the history a session carries, not a fixed number like 200K.
• How much do connectors and plugins add to every turn?Every connected tool loads its full description into every turn. On one Mac that was 126,000 tokens before any work. Here is how to measure and cut it.
• Claude Code subagents: what they are and what they costWhat Claude Code subagents are, when to use them, and what they cost. Each carries its own context; on one heavy day they made 61% of all turns.
• Where does Claude Code keep its logs?Claude Code writes a log file per session under ~/.claude/projects. Here is what each line holds, where subagents go, and two traps when adding up tokens.
• How to see Codex token usage by sessionCodex records token counts for every reply in ~/.codex/sessions. Here is how to see which Codex session used the most, and how it differs from Claude Code.
• Your agent is waiting on a check that already failedAn agent that opens a pull request and waits can sit for hours after a check goes red. Here is how to get told the moment it happens.

Comparisons

• Cost Doctor vs ccusageccusage is a free command line tool that reports token usage. Cost Doctor is a Mac menu bar app that says which session is the problem and why.
• Cost Doctor vs CodexBarCodexBar is a free menu bar app that shows how much of each provider's limit is left. Cost Doctor shows your limits too, plus which session used them and why.
• Cost Doctor vs HeadroomHeadroom compresses what your coding agent is sent, for a monthly fee. Cost Doctor shows which session is costly and why, for a one-time price.
• Cost Doctor vs SessionWatcherSessionWatcher tracks usage and rate limits across many AI coding tools. Cost Doctor shows limits for Claude Code and Codex, why a session is costly, and failed checks.