Why spending limits matter
AI agents consume credits when they run. The safest way to control spend is to set the cap at the agent level before the workflow is allowed to scale.
Pinksheep already gives you two useful control points: an estimated credit range in the agent plan and a monthly spend cap on the agent detail page. That is enough to set a safe starting boundary without inventing a larger budgeting system than the product currently proves.
Implementation steps
1. Start with the estimated credits per run
Before deployment, review the plan the builder generates. The manifest shows an estimated credit range per run. Use that as your starting input rather than guessing from generic AI cost advice.
A simple way to choose the first cap is to multiply the estimated credits per run by the number of runs you expect in a month, then round conservatively.
2. Set the monthly spend cap on the agent
Open the agent detail screen and use the `Monthly spend cap` field. The product copy in that screen is explicit: leave it blank for no cap, and runs stop once the monthly limit is reached.
If the workflow has uncertain volume, start lower and raise it after you observe real usage.
3. Run once and watch real usage
The safest first test is a real run after the cap is set. That gives you a better feel for whether the estimated range and the actual workflow line up.
4. Adjust after the first real month
Once you have real usage, tighten or raise the cap based on what the workflow actually costs. The point is not to predict the perfect number on day one. It is to avoid uncapped drift while you learn the workflow.
Limit models to use today
Monthly cap per agent
This is the clearest proven control today. Set a monthly cap on the agent itself and stop new runs once the limit is reached.
No-cap mode
Leaving the field blank removes the cap. Use this only when the workflow is well understood and the owner accepts the trade-off.
Conservative launch cap
Use the estimated credits per run to pick a cautious starting number, then revise it after the first real month.
Manual raise after real usage
Treat the first cap as a starting boundary. Raise it only after you have a clean month of real runs and a clearer sense of what the agent actually costs.
Best practices
- Start with the plan estimate. Use the estimated credits per run in the manifest as your first budgeting input.
- Use the monthly cap field on the agent.This is the clearest proven spend control in the product today.
- Leave blank only on purpose. No cap is a real option in the UI, so treat it as an explicit decision, not the default.
- Adjust after real usage. The first cap is a starting boundary. Tune it once the workflow has run for real.
- Do not overclaim what the cap does. Keep the language to a monthly run-start limit, not a per-step or per-tool hard stop.
Frequently asked questions
What happens when an agent hits its spending limit?
The current safe claim is simple: runs stop once the monthly limit is reached. Treat the cap as a monthly control on whether a new run can start, not as a more advanced budgeting system than the product currently proves.
How do I choose a starting spend cap?
Start with the estimated credits per run you see in the plan, then multiply by expected monthly run frequency. Begin conservatively, watch actual usage, and adjust after the first real runs.
Can I leave the spend cap blank?
Yes. The agent detail screen supports leaving the cap blank for no cap. Use that deliberately. If the workflow has unpredictable volume or higher risk, setting a monthly cap is the safer default.
Does the spend cap stop a run mid-way?
Not as a per-step control. Current product truth supports a monthly spend-cap check before new runs start. Do not describe it as a per-step or per-tool hard stop.