Skip to main content

Spend Caps

A spend cap limits how much your organization’s agents can spend on AI models in a day, a week, or a month. When a cap is reached, new agent runs are blocked until the period resets or an admin raises the cap. Caps guard against runaway usage: a scheduled task that fans out more than expected, a long multi-step job that keeps going, or a busy week nobody planned for. No caps are set by default, so nothing changes until you add one.
Settings → Spending limits: daily, weekly, and monthly spend caps

Set a cap

Only Organization Admins (org:admin) can set or change spend caps. See Roles & Permissions.
  1. Go to Control Hub → Settings → Spending limits.
  2. Enter a dollar amount for any of Daily cap, Weekly cap, and Monthly cap. Leave a field blank for no limit in that period.
  3. Click Save.
You can set one, two, or all three. A run is blocked if any cap is reached. Under each cap, a bar shows how much has been spent so far in the current period. It turns amber at 80% and red once the cap is reached. A dot next to Spending limits in the Settings menu shows the same warning from any other settings page.

Per-user caps

Per-user caps limit how much each person can spend, so one busy member can’t use up the whole organization’s budget. They work alongside the organization caps: a run is blocked if the organization or the person has reached a cap.
  1. Go to Control Hub → Settings → Spending limits and scroll to Per-user caps.
  2. Enter a standard daily, weekly, or monthly cap. It applies to every member, admins included.
  3. To give someone a different limit, find them under Members and click Set custom limit. Their custom limit starts as a copy of the standard. Change the amounts, then click Save.
A custom limit replaces the standard one for that person. A period left blank in a custom limit means no cap for them in that period, even if the standard sets one. Click Use standard to remove a custom limit. The Members list shows each person’s spend in the current period against their limit, with the people closest to their limit first.

Whose spend counts

A person’s spend is everything in the conversations they own, the same as the per-user cost in Control Hub → Analytics. That includes:
  • Chats they start
  • Their scheduled tasks
  • Sub-agents called from their conversations
  • Agent runs started by email they send
In a conversation shared with other people, every turn counts toward the person who started the conversation. If that person has reached their cap, the conversation stops for everyone in it. A custom limit stays in place after someone leaves the organization (they’re listed as Former member), since their scheduled tasks can keep running.

When caps reset

Periods follow the calendar in UTC:

What happens when a cap is reached

  • Chats: the conversation stops with a message saying which cap was reached and how much has been spent. The message people sent is kept, so nothing is lost.
  • Scheduled tasks: the run is skipped and recorded as failed in the task’s run history, with the same message.
  • Sub-agents: an agent that calls another agent gets the message back as an error.
  • Work in progress: a long-running task stops at its next checkpoint instead of continuing. The stretch of work already underway (up to a few minutes) finishes first, so spend can go somewhat past the cap.
To unblock your organization, raise or clear the cap, or wait for the period to reset. When a per-user cap is reached, the message tells the person it’s their own cap. An admin can give them a custom limit, or they can wait for the period to reset.

How spend is measured

Spend is estimated from the tokens each agent uses, priced at each model’s list price — the same figure as the cost in Analytics and on each person’s Usage page. It is not your provider’s bill, which can differ slightly because of discounts or rounding. The estimate covers agent conversations, scheduled tasks, sub-agents, and app actions. It does not include:
  • Knowledge base processing (embedding and extraction)
  • Conversation titles
  • Models without a list price in Aster, such as some custom Azure deployments
For a hard limit on your provider bill itself, also set a spend limit in your AI provider’s console (for example, a workspace spend limit in the Anthropic Console or a project budget in the OpenAI dashboard). Aster’s caps stop runaway agent work early; your provider’s limit caps the bill.