# Claude Code usage in terminal: read it at a glance

> The command is easy. Remembering to run it before a limit stops you is the part worth fixing.

URL: https://claudegpt.si/news/claude-code-usage-in-terminal-glance  
Published: 2026-11-18T16:22:02.845Z  
Author: ClaudeGPT Si Newsroom  
Section: Guides  
Topics: Claude Code, AI limits, Alerts, Developer tools  
Publisher: ClaudeGPT Si

## Key takeaways

- Claude Code usage in terminal comes from /usage, which shows session tokens, plan bars, a cost estimate and recent activity.
- The prompt view has to be opened and remembered, and it covers one machine at a time.
- A desktop tracker can hold the supported windows in view and send one alert per reset cycle at a threshold you choose.

Claude Code usage in terminal is one command away: /usage prints session tokens, plan bars and a cost estimate without leaving the prompt. Reading it takes ten seconds. Remembering to read it before a limit ends your afternoon is the harder part, and that is where a desktop view and a threshold alert earn their place.

## The check nobody runs at 4pm

Picture a Thursday. You have two sessions open, the first one has been alive since morning, and a refactor is half finished. The question is not how many tokens you have used in total. It is whether there is room to finish. /usage answers that in a glance, then disappears when you press Escape. Six turns later you have forgotten the number.

## What the prompt shows

The Session block gives total cost, API and wall duration, code changes, and a line per model with input, output, cache read and cache write counts. A prompt cache line follows once a response has landed. On a paid Claude plan the same screen adds usage bars and a breakdown of what is consuming your allowance, and d or w switches between the last 24 hours and the last 7 days. /cost is an alias, so older habits still work.

## Three ways the glance fails

It is pull, not push. Nothing taps you on the shoulder at 80 percent.

It is per machine. Sessions on another laptop, and anything you did on claude.ai, are outside the figures.

It is short. The session block resets when /clear starts a new session, and the local files behind the screen are deleted after cleanupPeriodDays, 30 days by default.

| What you want | Where to look | Catch |
| --- | --- | --- |
| Tokens this session | /usage, Session block | Resets on /clear |
| Plan headroom | /usage bars | One machine, approximate |
| A warning before a limit | A tracker with alerts | Has to be set up once |
| Last month | A tracker with history | Not in the CLI at all |
| A second tool's spend | That tool's own screen | Each one differs |

## Turning a glance into a ping

[UsageAtlas](https://usageatlas.com) is a free, open-source desktop app that keeps the supported Claude Code windows in view with remaining capacity, the reported reset time and a history of tokens and estimated cost. You set a threshold, and it sends one desktop alert per reset cycle, so a busy afternoon produces a warning rather than a stream of them. It runs on Windows, macOS and Linux, needs no account, and does not read prompts or responses. An optional free account saves history.

The point is not to stop opening /usage, but to have a second signal for the moments when you do not.

## Where a page watcher fits instead

AyeWatch watches public pages, topics and social accounts for changes, which is a different job from reading your own token counts. Keep it for the pages around your work, such as a provider's pricing or status page, and let a local tracker answer the usage question. Using one tool for both would mean trusting a page monitor with numbers it never sees.

## FAQ

### Does /usage show the reset time?

The screen shows your plan bars and recent activity. When a limit actually stops a request, Claude Code names the window that ran out and when it resets, and /rate-limit-options lists ways to keep working.

### Is one alert per reset cycle enough?

For most people, yes. The threshold is the control: set it where you still have room to change course, not at the point where the next request fails.

### Will a tracker work with an API key instead of a subscription?

UsageAtlas reads supported local activity and sign-in information, so it reports what your account provides. Local token and cost history works either way, and plan windows appear when the account reports them. Cost figures stay estimates.

## Frequently asked questions

### How do I see Claude Code usage in terminal?

Type /usage at the Claude Code prompt. The Session block shows this session's tokens and an estimated cost, and paid Claude plans add usage bars and a breakdown of recent activity.

### Can I get a warning before I hit a limit?

Not from /usage, which shows the state when you open it. UsageAtlas can send one desktop alert per reset cycle when a supported window crosses the threshold you set.

### Does a usage tracker sit between Claude Code and Anthropic?

UsageAtlas does not. It reads local logs and your existing sign-in, so nothing is added to the request path.

## Sources

- [Track Claude Code usage, limits, cost, and history](https://usageatlas.com/guides/claude-code-usage-tracker), UsageAtlas
- [Get an alert before you hit an AI coding limit](https://usageatlas.com/guides/ai-coding-usage-alerts), UsageAtlas
- [Manage costs effectively](https://docs.claude.com/en/docs/claude-code/costs), Anthropic
