This seems like a ridiculous amount of credits for a review that produced very little output. It’s worse than just prompting directly and uses way more credits. I just accidentally clicked “review again” which I canceled and now my 38 credit review from before is gone. Is there a correct way to use this feature?
I’ve sent you a PM!
Hey @johnfewell! Moving the convo back here.
Agent Review runs multiple AI passes over your code changes, including a bug-detection pass and false-positive validation to filter out noise, among others.
This is by design to reduce false positives. The 38 requests you saw represent the total cost of individual model calls.
If Agent Review found a few issues, that should be a good signal that your code is solid.
That’s good feedback. I’ll pass it on!
Right, but this pricing doesn’t make sense: 2.4M tokens is one credit, and a subpar code review is 38 credits using 781K tokens? Coderabbit is much more cost-effective at this level. The review feature is broken. I would encourage everyone to prompt directly and disable the agent review feature.
Dear @Colin , running into a similar issue: Agent Review requests have been retro-actively increased to crazy amounts. I didn’t understand why I kept getting messages to enable usage-based pricing until I checked the log and noticed that I reached my 500 request-limit even though I barely used Cursor this month. These request amounts is aburd - especially when it sometimes fails and doesn’t even return results
