# Usage per machine > The gauges of the Usage panel show the quota of your ACCOUNT, as Claude or Codex publish it: every computer, claude.ai and any other tool together. - Area: Desktop, mobile and web - Plans: All plans - Last checked against the product: 2026-10-22 - Web page: https://agentsroom.dev/docs/usage-per-machine ## What it does The gauges of the **Usage** panel show the quota of your ACCOUNT, as Claude or Codex publish it: every computer, claude.ai and any other tool together. **Per machine** shows the other side: what each of your computers spent, session by session. Each desktop reads the transcripts its Claude Code and Codex CLIs keep on disk (including the sessions you ran in a plain terminal, outside AgentsRoom), counts the tokens of every session per day and per model, prices them at public API rates, and sends the figures to your account so your other computers and your phone see them too. You get, for today, 7 days or 30 days: an estimated cost, tokens and a number of sessions, the cost per machine, per day, per CLI and per model, and the list of sessions with their project, their agent ("Outside an agent" for a terminal session), models, start time, duration, tokens and cost. ## Where to find it - Desktop: footer **usage** badge > **Usage** panel > in its title row, the **Account** / **Per machine** switch. **Per machine** replaces the gauges of the left column; the right column (overrides, quota rules, insights) stays. - Above the figures: **This machine**, one chip per other computer of the account (its host name), **All machines**, and the period (**Today**, **7 days**, **30 days**). With **All machines**, a **By machine** list compares the computers; click one to open it. - The refresh button of the panel reads this machine's transcripts again before reloading. - Phone: **Activity** tab > **Usage** segment > **Account** / **Per machine** at the top. ## How to use it 1. Open the **Usage** panel and pick **Per machine**. It opens on **This machine** over **7 days**. 2. Pick another computer, or **All machines** to compare them. 3. Read the session list: the newest first, the 50 most recent of the period. The amounts are estimates at public API prices, not a bill: a subscription is not charged per token. "At least" before an amount means a model of the period has no known price, so its cost is missing. ## Settings None. The figures are sent by every signed-in desktop about every ten minutes (and when the panel opens), whatever the data-sharing switch says: they are your own data, shown back to you, like the project statistics. ## Agent tools (MCP) None. ## Providers Claude Code and Codex, the two CLIs whose token counts AgentsRoom reads from disk (every Claude account directory signed in on the machine, every Codex home). Sub-agent turns count in the session that launched them. Sessions of the other CLIs (OpenCode, Antigravity, Grok, Cursor, Aider and others) are not counted. ## Mobile Present, read only: the **Per machine** view of the **Usage** segment shows every computer of the account (there is no "This machine" on the phone, which has no CLI transcripts), the same periods, figures and session list. It reads the account directly, so it works without a desktop connected. ## Limits - Only what a computer's transcripts still hold: a CLI that deleted old transcripts, or a computer that never ran a recent AgentsRoom, has nothing to send for those days. - Days are the calendar days of the computer that ran the session. Figures are kept 90 days on the account; the longest period shown is 30 days. - The first read of a month of transcripts on a heavy machine takes a few seconds in the background; later reads only look at what changed. - macOS, Windows and Linux. ## Common questions - **Why does my machine's cost not match the account gauge?** The gauge is the provider's percentage for the whole subscription (all computers, claude.ai, other tools); the machine view counts only the CLI sessions each computer ran, priced at public API rates. They measure two different things. - **Are sessions I ran in my own terminal counted?** Yes, for Claude Code and Codex: they write the same transcripts, so they appear as "Outside an agent". - **Is my code or are my prompts sent?** No. Only counts per session and day: tokens, estimated cost, model, the project name, the agent name and the computer's host name. Never a prompt, a path or the CLI session id. - **Another computer is missing.** It appears once it ran a recent AgentsRoom signed in to the same account and had Claude Code or Codex activity in the period. ## Related - [Claude Code Token Usage](https://agentsroom.dev/docs/claude-code-token-usage.md): the Usage panel, its account gauges and the per-session token meter. - [Project Statistics](https://agentsroom.dev/docs/project-statistics.md): time, prompts and cost per project and per agent.