Skip to main content
TripleBits Apprentice - Software for Agents

Menu

Sign In Register

Configure Agent Budgets And Cost Controls

Set Apprentice agent budgets, Free Agent mode, daily weekly or monthly limits, pause or notify behavior, and current spending review.

18 views

Configure Agent Budgets And Cost Controls

Budgets help keep agent spending visible and bounded.

Use budgets for agents that call paid API providers, subscription-backed services, or external tools that can create cost. Local or offline model runtimes may not have meaningful token cost, but you can still use other runtime limits such as max duration and tool-call turn limits.

Open Budget Settings

Open the agent, then select Config > Budget.

The Budget tab shows:

  • Free Agent toggle.
  • Budget period.
  • Budget amount in USD.
  • Action on limit.
  • Current spending.
  • Reset time.

Free Agent

Free Agent skips budget enforcement for that agent.

Use this for local or offline model runtimes, such as LM Studio or Ollama, when token cost is effectively zero for your setup.

Do not use Free Agent for paid APIs unless you intentionally want no budget enforcement.

Set A Budget

Select Set Budget, then choose:

  • Period: Daily, Weekly, or Monthly.
  • Budget: amount in USD.
  • On limit: Pause execution or Notify only.

Use small budgets while testing a new agent.

Choose The Limit Action

Pause execution stops the agent when the budget limit is reached.

Notify only continues execution but warns you when spending reaches the limit.

For unattended schedules or integrations, use Pause execution unless you are comfortable reviewing cost after the fact.

Review Current Spending

The Budget tab shows current spend against the configured limit.

Watch the percentage indicator:

  • Under 60% is normal.
  • Around 80% deserves review.
  • At or above 100% means the budget is exceeded.

The reset time shows when the period rolls over.

Budget Practical Defaults

For a new paid API agent:

  • Start with a small daily budget.
  • Use Pause execution.
  • Run one manual test.
  • Review Activity and Run Detail.
  • Increase the budget only after the agent behaves correctly.

For a scheduled agent:

  • Use a budget period that matches the schedule cadence.
  • Keep max duration low enough to prevent long accidental runs.
  • Avoid broad prompts that make the agent inspect too much.

For a local model agent:

  • Mark it as Free Agent if cost tracking is not useful.
  • Still use duration, permission, and capability limits.

Troubleshooting

If an agent stops unexpectedly, check whether the budget limit was reached.

If the Budget tab says the agent is Free, toggle Free Agent off before setting a budget.

If spend is higher than expected, review Run Detail for token usage, provider choice, tools used, and duration.

If a scheduled or integration-triggered agent spends too much, disable the trigger first, then adjust the prompt, model, budget, and permissions.

Next Step

After budgets are set, review provider and model choices for cost and reliability.