How do you track Codex CLI usage?

From inside a session. Codex reports on the account it is signed in to:

On Plus and on a Standard Business seat, Codex usage is counted per five-hour period, and weekly limits may also apply. Pro plans currently have no five-hour limit, and neither does a Business Premium seat.

OpenAI publishes estimates rather than fixed message counts, because usage depends on the model and the task. Local messages and cloud tasks share one allowance, and ChatGPT Work draws on it too.

Enterprise and Edu workspaces on flexible pricing have no fixed rate limits. They spend credits instead.

The turn already running is allowed to finish, subject to fair use. After that you can wait for the reset, buy credits on a plan that sells them, or run Codex on an API key at standard API rates.

A Business workspace has workspace analytics, with a usage table per member: seat type, credits spent and messages sent. Its Codex analytics add active users, credits and tokens, threads and turns, and a user leaderboard.

Enterprise workspaces with Codex enabled also get an Analytics API, for loading the same reporting into your own systems. Business owners and admins can set monthly credit limits per seat type, and override them for one person.

/status and /usage describe one person's account, so a team total means asking everybody. Workspace analytics cover one ChatGPT workspace, and Claude Code is not in it.

Codex run on an API key is billed on OpenAI's API platform, which a workspace's usage controls do not govern. A credit limit per person applies to credits spent after a seat's included usage, not to the included usage itself.

An owner connects the company's OpenAI accounts once: a ChatGPT subscription account, an OpenAI API key, or both. Each developer signs in with selan login and runs selan codex instead of codex, and every argument passes through. No OpenAI key lands on the laptop. A bare codex fails closed rather than spend a company account.

Every request is recorded with who ran it, which account served it, which model answered and what it cost. Usage, in the Selan control app, draws Codex as its own band beside Claude Code and Claude Desktop, over the last 24 hours, 7 days or 30 days, in UTC. Filter by client to see only Codex, and the ranked list shows who used it and in which team.

A subscription seat is priced at what the same requests would have cost per token, next to what your API keys actually cost. The setup is in the Codex docs.

Yes, with selan usage. The statusline Selan adds belongs to Claude Code, and Codex gets none from Selan.

selan usage prints the weekly window of the company's ChatGPT accounts, and the five-hour window where the plan reports one, each with its reset time once one is due. A window that is not reported prints as not reported, never as 0%. If an owner set the developer a spend limit, the same output shows it as spent of limit.

What the Claude Code side shows is in the statusline docs.

Tracking only shows who spent too much once they have spent it. In Selan you can set a spend limit per person, by day, week or month, and it follows them across every pay-per-use key you added, an OpenAI API key included. It binds the moment it is reached, and the refusal says which limit it was and when it lifts.

When the company has a model on more than one account, a pick from the list falls through to the next account if one is out of quota.

Limits are on the spend page, and the same figures for Claude Code in Claude Code usage per developer.

Free for up to three seats, and the Free plan needs no card.