Hit 10,000 API calls in one day on my side project and felt sick

I run a little trading alert bot with a language model reading news headlines, and I checked my usage page on Tuesday and saw 10,247 calls in 24 hours. That was not a spike from a big news day, that was just my retry loop hammering the endpoint every time the model gave me a weird response. I sat there doing the math on what the next bill looks like and it was over $40 for one day of a bot that 12 people use. Turns out a rate limit error from the provider was making my code resend the same prompt up to 8 times, and no log ever showed it because I only logged successful runs. How are you all catching silent retry loops like that before the bill shows up?

1 comments

Log in to join the discussion

Log In

1 Comment

the_logan25d ago

The line "I only logged successful runs" is the part that got me, because I used to think logging failures was enough and that was just noise. Then I watched a buddy burn $200 in a weekend on a summarizer that kept retrying on 429s, and his logs were full of happy green checkmarks the whole time. What changed my mind was setting up a simple counter that fires an alert if the same prompt hash shows up more than twice in five minutes. Two weeks later it caught a loop where the model kept returning an empty string and the code just kept asking again. Now I treat retries as something to count, not something to hide.

1