The cap is why we were willing to turn AI on at all.
Shipped in 3.13.2. If you bring your own AI key, you can now cap what a flow is allowed to spend, and AI output passes a moderation check before it reaches a customer.
What is new:
- A spending cap per key, so an unexpected loop cannot run up a bill on your
provider account.
- Anomaly detection on hourly spend, which pauses a key whose usage jumps well
outside its own recent pattern.
- Output moderation on generated replies before they are sent.
A note on how AI works here, because it comes up: ChatinFlow does not resell model access. You add your own provider key, the calls go to your account, and you see the usage on both sides. That is why the caps live on the key rather than on a plan.
The moderation step is not optional and not configurable, deliberately. A generated reply that reaches a customer is your brand speaking, and the failure mode is worse than a missed message.
2 Answers
Does the anomaly pause notify, or do you find it in the key list? ---
Sign in to reply and save your draft.