How to Cap AI Spend on a Webhook Across Zapier, Make, and n8n

When a webhook calls a paid model on every request, a sudden spike or a runaway sender can cost real money fast. The defense is a simple daily counter plus disciplined prompting. The same idea works in Zapier, Make, and n8n; only the storage step differs.

What you need

• An AI webhook already running in Zapier, Make, or n8n
• A counter store: Storage by Zapier, a Make Data store, or n8n static data
• A daily request limit you are comfortable paying for

Step 1: Read and bump a daily counter

At the start of the workflow, read a counter keyed by today's date, for example calls-2026-06-21. If it is at or above your limit, stop before the model call. Otherwise increment it and continue.

const day = new Date().toISOString().slice(0, 10);
const data = $getWorkflowStaticData('global');
const key = 'calls-' + day;
data[key] = (data[key] || 0) + 1;

if (data[key] > 500) {
  throw new Error('Daily AI limit reached - skipping model call');
}
return $input.all();

Step 2: Shrink the prompt and output

Cost scales with tokens. Trim the input to only what the model needs, set a low max_tokens, and pick a smaller model when the task is simple. A classification job rarely needs a flagship model.

{
  "model": "gpt-5-mini",
  "max_tokens": 60,
  "messages": [
    { "role": "user", "content": "{{trimmed_input}}" }
  ]
}

Step 3: Alert when you near the cap

Add a branch that sends a notification when the counter crosses, say, 80 percent of the limit. You find out before the cap kicks in rather than after requests start failing.

Result: even a flood of webhook traffic stops at your chosen daily ceiling, so your AI bill stays predictable no matter what hits the URL.