I made ai-usage because I kept hitting the Claude Code 5-hour limit in the middle of a task. With a couple of laptops, a build server and some bots on shared logins, I never knew who was using what.
ai-usage is one small binary for macOS and Linux (Windows is experimental). Every 15 minutes the system scheduler runs it: it reads what Claude Code, Codex, Grok and Hermes already write on disk, asks the installed tools for quota, and exits. No daemon, no account, no config file. It doesn't call the provider APIs or read credentials, so it spends none of your quota.
What you get:
- Every subscription on one page: 5-hour, weekly and per-model windows, with weighted tokens for today and the last 7, 30 and 90 days.
- A forecast to the reset from your own pace (over, tight, ok, under), and when a window will run out.
- An attention list that appears only when something is wrong: a used-up limit, a machine that went silent, an old release, an idle paid subscription.
- Utilization history: how full each window got before each reset over the last 90 days.
- Team mode: up to 50 laptops, servers and bot containers share one team key, with each machine's share of each subscription and tokens per git repo.
- A Mac menu bar app (macOS 14+) with rings, details per subscription and notifications.
- JSON output for scripts and agents.
Privacy: prompts, replies, code and file contents are never kept or sent. Names, emails and project paths are sealed with the team key; the relay (the free project relay, or your own) sees only tools, plans, percentages, reset times and token counts.
Free and open source under the MIT license. Forecasts are estimates from your own pace.
Sergey Baus added ai-usage as alternative to ccusage, Claude Code Usage Monitor, Usage4Claude, ClaudeBar and 4 more
No comments or reviews, maybe you want to be first?