AI agent metering that stops a run at the cap

The same run, with a meter

You change the model call, not the agent. The rules live in the Console.

1. Name whose budget the run spends

   Replace the provider call with usage.chat and pass the customer's identity and a workflow name. Use the same identity you use in Stripe.

2. UsageFlow checks before every call

   It reads the customer's running total and applies the policy: allow, route to a cheaper model, degrade, or block. Prompts still go from your server straight to the provider.

3. Handle the stop like any other result

   A blocked call raises a rejection error and the provider is never contacted. Catch it and end the run with a clear message.

   TypeScript Copy

   import { usage, VibeRejectionError } from '@usageflow/vibe';
   
   async function runAgent(customerId: string, task: string) {
     const messages = [{ role: 'user', content: task }];
   
     for (let step = 0; step < 12; step++) {
       try {
         const turn = await usage.chat({
           identity: customerId,         // whose budget this run spends
           workflow: 'research-agent',   // the policy that sets the cap
           provider: 'anthropic',
           model: 'claude-sonnet-5',
           messages,
         });
         messages.push({ role: 'assistant', content: turn.content });
         if (isDone(turn)) return turn.content;   // your own stop condition
       } catch (err) {
         if (err instanceof VibeRejectionError) {
           return 'This run reached your usage limit.';
         }
         throw err;
       }
     }
   }

4. Bill what ran

   Add a Stripe meter effect to the policy and each call's usage is reported to your Stripe meter. What you limit and what you invoice come from the same numbers.