Check AI account usage

Before giving an agent a long task, check the allowance on the account it will use. ThinkingMach shows what the provider reports when you ask; it does not estimate a subscription balance from your task costs.

Usage checks and AI connection setup do not require an experimental setting.

Check a saved account

  1. Open Connectors and the saved AI account.
  2. Find Usage and select Check usage.
  3. Read the returned limits and reset times. Select Refresh when you need another observation.
Account Supported check
Anthropic subscription Provider-reported Claude usage windows and overage when available
OpenAI subscription Provider-reported allowance windows, reset times, and overage when available
Grok subscription Provider-reported subscription limits and overage when available
OpenRouter API key Key credit limits, reset cadence, and free-model limits when available
Other API-key methods Unavailable for this sign-in method.

The check uses the stored connection you selected. It does not probe an ambient CLI login. Opening an account page does not trigger the check automatically.

Read the result

A limit may show a percentage, a used amount and cap, or a remaining allowance. Limit reached means the provider reports the limit is exhausted. Blocked means it explicitly reports usage is not allowed. Reset time or cadence appears when supplied.

Missing fields show Not reported, never zero. Grok may omit included-plan usage, and some Claude credentials can run models without permission to read usage. Usage access denied. does not mean your allowance is exhausted.

Overage describes the provider's extra-usage state. On · Availability unknown means enabling overage alone has not established a funded usable balance. The observation timestamp tells you when the response was read.

Checking usage does not refresh your credential, buy credit, change the account selected for a task, or stop an active run. It is a provider observation, separate from ThinkingMach budgets and per-run cost accounting.

See quota on the Costs page

The Providers tab on the Costs page also shows a Subscription quota section for Anthropic and OpenAI subscription accounts. It lists each connected subscription account you're allowed to use, under its own name, and reads it with that account's stored credential — not a CLI login on the machine running ThinkingMach. Unlike Check usage, it refreshes on its own every few minutes while you're on the Providers tab.

If a refresh fails briefly, the last good reading stays on screen. If the account needs to be connected or reconnected, the reading is cleared and the section asks you to reconnect it here in AI connections.

See what a subscription costs per month

Usage checks tell you how much allowance is left; they don't tell you what the plan costs. For that, use the Subscriptions tile at the top of the Costs page. It lists the monthly fee for each subscription account your agents use, counts a shared seat once, and lets you enter the real price when ThinkingMach can only estimate it — or can't tell at all.

Repair sign-in from a task

When a task cannot start because its AI credential is missing or expired, it shows an authentication connection card. Open that card and connect your account or reconnect the existing one. After the card finishes successfully, the task resumes automatically through normal admission.

Only the account owner can reconnect it. If the task uses someone else's account, the card explains who must restore it. If that account is no longer available to you, ask its owner to restore access or choose another available connection in the agent's settings.

A reconnect retains the existing account's audience, agent access, and default selection. It does not broaden sharing. In agent settings, Reconnect account repairs your unavailable personal default; Connect an account… lets you establish another account.

Troubleshooting

Message What to do
Sign in again to check usage. Reconnect the stored account.
Usage access denied. The provider denied usage-read access for this credential. Check usage at the provider if necessary.
Too many checks. Try again later. Wait before another usage check. This does not consume your model allowance.
Reconnect to check usage. Restore the connection before checking again.
Provider unavailable. Try again. Retry later; the provider did not supply a usable response.
Couldn’t read usage. Try again. The response could not be read. Missing data is not treated as zero.

API equivalent

Board-authenticated users with company access can request the same observation:

GET /api/companies/{companyId}/ai-connections/{connectionId}/usage

Pass ?grantId={grantId} when the connection has multiple eligible credential grants. The response includes checkedAt, status, provider-reported limits and overage, and an errorCode when unavailable. You can only read a credential you are allowed to use.