Lighter, with costs you can trust and a 30-day usage view.
- Usage window. Choose Usage… in the menu bar menu to see 30 days of tokens and API-rate cost by day, model and project. It reads your transcripts only while the window is open.
- Costs counted once. A resumed Claude session no longer bills the history it copies, and subagent usage counts toward the session that started it. Rows read "≈ $41 API" so nobody mistakes them for a bill.
- Lighter. Muster reads only what changed. With ten busy agents it averages about 0.4% of one core and about 20 MB of memory.
- Answered elsewhere, gone here. Answer a prompt in Claude's own dialog and Muster drops it right away.
- This week. Settings counts the prompts you answered without switching and the jumps you made.
- Safer hooks. Muster keeps your original settings file as
settings.json.muster-original, leaves symlinks and permissions as they are, and redoes its edit if Claude writes the file at the same moment.
Codex users: in Settings, under "Answer prompts from Muster", turn Codex off and on once to pick up the shorter wait, then run /hooks in Codex to trust the updated hook.