Budgets and Spending Limits

The story is simple: give your AI agent money, but not unlimited trust. 4Mica lets an agent pay for useful services. Your policy decides how much the agent can spend and under what conditions.

​

Core limits

Start with hard limits. Your policy should define the largest amount an agent can spend in one request, across one task, and over a time window. It should also constrain where that money can go: which sellers, categories, networks, and assets are allowed. Set the first budget caps from dashboard, then raise them only after the agent behaves predictably.

​

Fixed and usage-based prices

Before paying, your agent should know what kind of price it is accepting. Some sellers charge a fixed amount for each request or task. Others charge by usage, such as API call, token, second of compute, data unit, or completed result. Some work starts with a quote that expires if you do not accept it. Fixed prices are easier to approve automatically because the final cost is known before signing. Usage-based work needs stronger guardrails. If the final cost can grow, use a cap, staged payment, or human approval threshold.

​

Approval gates

Approval gates decide when your agent can keep going by itself and when it must pause for a person or higher-level policy. The most common triggers are payments above a per-request threshold, tasks nearing their budget, new sellers, high-risk routes, downstream paid agents, changed estimates, and attempts to continue after failure. Approve before signing the payment, not after the agent has already spent. If your approval flow needs custom thresholds or multiple reviewers, contact 4Mica support before production.

​

Stopping mid-task

Your agent should stop when continuing would break policy or waste money. That includes an exhausted budget, denied approval, repeated seller failures, suspicious payment requirements, user cancellation, or a task that can no longer be completed within budget. Stopping is a feature. It prevents a useful agent from becoming an uncontrolled spender, and it gives the user a clean point to approve more budget, change sellers, or abandon the task.

​

Hidden costs

Your agent may pay other services while completing one task. The visible seller price may not be the whole cost if the agent also buys model inference, data access, compute, storage, downstream agent work, retries, or premium latency. Show estimated cost before the task starts and actual cost after it ends. If the task changes, update the estimate before spending more. Use your dashboard history to compare estimated spend with actual downstream payments.

​

Refunds

Refunds are your policy. 4Mica provides payment evidence that helps both sides see what was paid. Your application should connect that evidence to task logs, seller terms, and support decisions. For your safety, prefer sellers with clear refund rules and strong delivery evidence.