I have no idea how I end up with this kind of billing where most of my instances are stuck or in cold status.
Hm… It looks like RAM is by far the largest contribution. Maybe some of the stuck sprites are using gigabytes of memory? (That might be why they’re stuck, actually, 
.)
You can email billing@fly.io to get more details (and possibly a partial refund, if it really was the stuck Sprites at fault).
https://fly.io/docs/about/support/#billing-and-account-support
Yeah anyway. I tolerated the stuck sprites because well it’s cheap no. Well it’s not anymore so I am cancelling this service. Good luck folks.
Most folks who post here are customers, and we’ll not implore you to stick around if you don’t want to. But do at least make an attempt at getting a refund - my impression is that Fly’s accidental costs policy is comprehensive and genuine.
Let’s see. From the response I don’t think I’ll be getting any refund. Getting charged over 200 dollars for RAM while the CPU charged is 15 dollars don’t make sense.
sprites are billed differently than fly machines or other compute. rather than having “packages” of, for instance, 2 CPUs and 4 GB RAM, a sprite lets you use as much cpu/memory as you want (up to the limit on your account) and charges you for however much of each resource you actually use.
now, I don’t know what you’re doing on your sprites so I can’t tell you whether it makes sense for you to be billed this amount; but I also don’t know our billing to be inaccurate so it’s likely that something in your sprites used as many resources as you are being billed for.
on the billing@ thread, an (unsolicited) bit of advice: if you say “I don’t understand my bill”, we are going to try to explain the charges on your bill. if you say “I want a refund”, and maybe mention accident forgiveness, we are going to look into getting you a refund. our support folks are generally very helpful but they can’t know what you want if you don’t explain it, and I assume this goes for most other places as well. 
Here’s the problem. We only use it for coding agents that often get stuck all the time and two instances of Hermes and OpenClaw that we barely used. Hence the charges for CPU is only 15 dollars.
Our expectation is that if the instances doesn’t do any work, it will go warm/cold. If CPU is barely used, I don’t understand how the RAM charges are more than 10 times of the CPU.
Do you mean a breakdown of e.g. mem usage per process or something of that sort? I think the best way to get this would be to run some monitoring yourself, within your sprite.
This is the case, it might be worth reviewing the things that can keep a sprite running: Working with Sprites | Sprites
RAM and CPU charges are down to the workload, so it’s certainly possible for RAM to be much higher than CPU. E.g. I use Sprites as a dev environment, and my charges for July are $44.53 for RAM and $5.13 for CPU. If I had to guess, I’d say agents generally use a fair bit of RAM, but most of the compute happens at the LLM provider, so the local CPU doesn’t really do much (think what tool calls the agent makes, it’s mostly file editing, file listing, etc.).
I got my last bill refunded. Thank you Profile - lillian - Fly.io for the tips.
