A simulated AI agent loop with a pre-flight spend check in front of every LLM call. Push the loop into a runaway and watch the guardrail trip — or hit the kill switch and freeze all spend instantly.
Hard cutoff blocks the call with HTTP 402 when the cap would be exceeded. Warn mode logs and allows it. The kill switch halts all spend immediately. The runaway loop simulates a stuck agent whose context window grows every iteration — costs accelerate until the guardrail trips.
| Model | Input | Output |
|---|
Illustrative pricing for demo purposes — not live provider rates.
| Time | Model | In tok | Out tok | Est. cost | Status |
|---|