ccusage alternative: the window, not the spend

Where ccusage is better

It tells you where the money went. Daily, weekly, monthly and session reports, with the models you used, and Claude Code's 5-hour blocks. CrewTower keeps no token or cost history at all; the one dollar figure it can show is the current Claude Code session's running total, with the beta Tell Claude plugin on.

It reads eighteen agents. Claude Code, Codex and sixteen others. CrewTower's usage meter covers Claude Code and Codex.

It is free and runs anywhere. Open source under the MIT license, straight from npx, on any system with a terminal, with no account involved.

It fits into your own tools. JSON output for scripts, and ccusage statusline to put its figures in Claude Code's own status line.

CrewTower vs ccusage

Where CrewTower is different: where the number comes from

ccusage builds Claude Code's 5-hour blocks from the project logs on your disk: a block starts with your first message and lasts five hours, and the limit it measures against is one you set, or your highest previous block. That is accurate for counting tokens, because the tokens are in the logs. It is a weaker basis for how close am I to being cut off, because your plan's limits live with your provider, and work from another machine or from claude.ai is not in those logs.

CrewTower answers that second question from the source. For Claude Code it calls Anthropic's usage endpoint with the login Claude Code already uses, about every five minutes, so the percentage in your notch is Anthropic's own figure rather than a reconstruction. For Codex it reads Codex's own rate limit reporting. The session window and the weekly window each show the percentage used and a countdown to the reset.

The other half of CrewTower

The usage strip is one line at the bottom of the panel. The rest of CrewTower is about a different problem: an agent runs unattended until it needs a human, and then it stops in a terminal tab you are not looking at.

CrewTower brings that moment into the notch with the command on the card, and you approve, deny or answer from there. The agent's hook is blocking on a local socket and your decision travels back down it, so the prompt resolves in place without your terminal coming forward. When you do want to be in the terminal, Jump lands on the exact tab or split pane.

If you searched for a ccusage alternative because you wanted a spend report in a nicer wrapper, that is not this. If you searched because you keep discovering, twenty minutes late, that something was waiting on you, that is exactly this.

Running both

Nothing conflicts. ccusage is a command you run when you want the breakdown; CrewTower watches sessions, answers prompts over the hook socket and keeps the window on screen. Plenty of people keep both, and the honest reason is that neither one answers the other's question.

Which one to pick

ccusage if the question is what you spent, on which project and which model.

A paid app for the same question: SessionWatcher puts spend and limits in the menu bar on macOS 14 and Windows, from $6.99 once for one tool until 31 October 2026 and $8.99 after, with a 7-day trial. It and the other usage apps are side by side in the usage monitors compared, and CodexBar is the free one for many providers.

CrewTower if the question is how much is left and which session is waiting on you.

Questions

Is it on Windows or Linux?

No. CrewTower is a Mac app and needs macOS 15 or later, on Apple silicon or Intel. ccusage itself runs on any of them.

Sources

Read on 19 October 2026: github.com/ccusage/ccusage for the reports, the eighteen sources, the MIT license and the release (20.0.37, 19 October 2026); ccusage.com/guide/claude for the local logs; the blocks guide for the 5-hour blocks and their limit; the live monitoring page for its removal in version 18; the statusline guide; the JSON output guide; sessionwatcher.com and its facts page for its prices, trial and platforms.