On this page

Group providers by meter

Subscription assistants such as Claude Code, Codex, Cursor, Kimi, Kiro, GLM, OpenCode, and Command Code need allowance and reset tracking. API services such as DeepSeek, Grok, MiniMax, and Perplexity need tokens, credits, and spend. Local runtimes such as Ollama and LM Studio need context, memory, queue, and speed.

GitHub Copilot adds premium requests, Devin adds ACUs, and Gemini-family tools may expose request quotas and local session statistics. Keep those units visible instead of converting them into a misleading universal percentage.

Review four questions every week

Which tool hit a limit? Which task caused the largest jump? Which subscription had substantial unused capacity? Which model produced the best accepted result for its usage? Those four questions are enough to change next week's routing.

Add plan cost only after the usage data is stable. One busy day should not decide whether a monthly subscription is valuable.

Keep your usage visible while you work. Get Super Notchy for $27 ↗

Use live and historical layers

The live layer should be small: current usage, reset time, and agent activity. Super Notchy provides that glance across supported providers at the edge of a Mac. The historical layer can stay in provider dashboards or a weekly spreadsheet with task outcomes.

This split keeps the daily workflow calm and the weekly decision evidence-based. The goal is not perfect accounting. It is fewer interrupted sessions and better choices about where the next task should run.

Sources and further reading