Usage and several accounts

Usage in the sidebar shows every plan your team's agents run on, and how much of each limit is used. The sidebar card on every page shows just one of your plans — the one for the tool you used most in the last 24 hours — with a "+N more" link here.

#Plan limits

One card per plan account, live: which tool (Claude, ChatGPT for Codex, …), whose it is, each limit window (for example "Session (5 h)" and "Week"), how much is used and when it resets. Bars turn amber from 80 % and red from 95 %. The most recently updated and most used come first.

A card whose limits have all reset with no fresh numbers since — for example an account from before you named it with AGENTOWER_ACCOUNT — is hidden while another card for the same tool and person has numbers.

• Claude limits come from the Agentower status line.
• ChatGPT (Codex) limits come from the Agentower hook for Codex: at the end of each Codex turn it reads only the limit numbers from that session's own record on your computer. Nothing to set up beyond connecting Codex.
• Other tools can send their limits through the API — see Plan limits for any tool.

#Several accounts

If you use more than one account for the same tool — a work and a personal Claude account, say — give each a name so they show separately. The name is yours; Agentower never uses the account's email or ID.

• Claude: add AGENTOWER_ACCOUNT=work to ~/.agentower/claude-code.env. If you run your second account with its own Claude settings folder (CLAUDE_CONFIG_DIR), put "AGENTOWER_ACCOUNT": "personal" in that folder's settings.json env block instead.
• Codex: AGENTOWER_CODEX_ACCOUNT=team in the same file (or AGENTOWER_ACCOUNT if you only use one label).

Names can use letters, digits, spaces and . _ -, up to 40 characters; anything else is left out. Each person can have up to 6 accounts; the least recently used drops off first.

#Tokens by model

Which model actually did the work: tokens and their API value today, over the last 7 days (today and the 6 days before) and this month, per model. Every plan sees this, Solo included. Claude Code's telemetry names the model that served each request, so a subagent running on Haiku counts as Haiku even when its session runs on Opus. Tools whose telemetry doesn't name the model (Codex, for now) count under the session's model. Days are UTC.

The same numbers feed the public stats counters byModelThisMonth, byModelAllTime and tokensThisMonth, and the tokens and cost in the Daily usage export.

#Tokens and API value

Under that, a table shows today's and the last 7 days' tokens and API value per person, tool and each session's main model, from each tool's own telemetry. API value is what the same tokens would cost on the pay-as-you-go API. On a plan such as Claude Max you pay your subscription instead — not this; it shows the value you get from the plan. Most tokens are re-reads from the tool's cache, which the API charges at about a tenth, so the token count is large and the value much smaller than it would be at full price.

#Who sees what

Everyone in the team sees the team's plans and usage, as on the board. Only the plans' percentages and reset times are kept — never prompts, costs from the tool's own bill, or account details.