Usage limits & budgets
Two kinds of controls shape what your instance can do and spend: plan limits set by your subscription, and budgets and settings you configure yourself.
Plan limits
Section titled “Plan limits”| Limit | Launch | Scale | Enterprise |
|---|---|---|---|
| Seats (people with an account) | 5 | Unlimited | Unlimited |
| Agents | 10 | Unlimited | Unlimited |
| Concurrent agent runs | 3 | 50 | By contract |
| Included storage | 50 GB | 500 GB | By contract |
| Backup retention | 7 days | 30 days | 90 days |
What happens at a limit:
- Seats — admitting a new person is refused until you remove someone or upgrade.
- Agents — creating a new agent is refused. Delete unused agents or upgrade.
- Concurrent runs — further runs wait in the queue and start as soon as a running one finishes. Nothing is dropped.
- Included storage — nothing is blocked; storage above the included amount is billed.
Model budgets per team
Section titled “Model budgets per team”Administrators set, per team, which models members’ agents may use and a spending budget over a rolling window of 1, 7 or 30 days, in Settings → People and teams → My teams → Models and budget. When the budget is reached, further model calls through the team’s policy are refused until spending drops out of the window. See Model gateway & budgets.
Per-agent controls
Section titled “Per-agent controls”In each agent’s Settings → General:
- Parallel capacity — how many tasks the agent may run at once (default 3).
- Timeout — maximum duration of one task (default 60 minutes, up to 120).
- Autonomy — scheduled and unattended work only runs when autonomy is on.
- Max turns in Guardrails — stops runaway tool loops.
Loops can also stop at a maximum number of runs or a maximum cost.
Per-creator agent quotas
Section titled “Per-creator agent quotas”Instances can cap how many agents each person may create by role (by default creator 10, operator 3, user 1; admins are not capped). Administrators adjust these in Settings.
Watching spend
Section titled “Watching spend”- Account dashboard — compute, storage and gateway usage for the current period.
- Instance — execution history with cost per run, per-agent cost and context usage, and model usage per team.
- Alerts — budget and cost alerts arrive in the Inbox.
Good habits: keep agents on the smallest profile that works, set timeouts, use loops with a cost cap, and turn autonomy off for agents you are not actively using.