Discussion
I've been using rool for a few weeks now and it's proving to be very good.
It's pretty quick, and it's ability to 'remember' previous queries is very helpful.
Over the course of a few weeks I was researching buying a used car. It was helpful narrowing down between different models, specs and options for different model years. It made a few errors cross-referencing previous queries, but once pointed out it corrected itself and hasn't made the same mistake twice. It made me more confident making my purchase and I'm happy with my decision aided by rool.
Same thing sizing footwear. Giving it examples of makes and sizes that fit well to compare with other shoes and lasts and it sized my new boots perfectly.
It summarised a finance agreement and clarified/confirmed a few things I was unsure of.
Basically, what I suppose most of them do, but it's Dublin based and all machines hosted in EU.
I'm only posting this as they seem to be a start-up of some kind and would benefit from more exposure. I'm not connected to them in any way, just a happy user that would like it to still be there tomorrow, and the foreseeable future.
click >> rool features webpage
It's pretty quick, and it's ability to 'remember' previous queries is very helpful.
Over the course of a few weeks I was researching buying a used car. It was helpful narrowing down between different models, specs and options for different model years. It made a few errors cross-referencing previous queries, but once pointed out it corrected itself and hasn't made the same mistake twice. It made me more confident making my purchase and I'm happy with my decision aided by rool.
Same thing sizing footwear. Giving it examples of makes and sizes that fit well to compare with other shoes and lasts and it sized my new boots perfectly.
It summarised a finance agreement and clarified/confirmed a few things I was unsure of.
Basically, what I suppose most of them do, but it's Dublin based and all machines hosted in EU.
I'm only posting this as they seem to be a start-up of some kind and would benefit from more exposure. I'm not connected to them in any way, just a happy user that would like it to still be there tomorrow, and the foreseeable future.
click >> rool features webpage
I would be quite hesitant to believe half of what they have put in their FAQ. While they might (but it would be a huge investment even if they use AWS etc for it), it is much more likely they are not hosting the models and are actually just using OpenRouter or Nano-GPT (or other) on their backend.
The fact that they offer 'credits' rather than explain how many tokens you get, and do not indicate which 'frontier' models they have available just tells me some of this is, frankly, BS.
The only frontier models they will be able to host themselves are the Chinese ones as well, Anthropic, OpenAI etc do not make theirs available to 3rd parties.
For example, Deepseek V4 Pro needs around 960GB VRAM to run at 4bit quant. Kimi K3, at Q1 still needs around 600GB of VRAM - https://unsloth.ai/docs/models/kimi-k3 and 1.56TB at Q8. These are frontier models that are available for 'self hosting'.
The start up costs, for this, are staggering. Which is why I suspect heavily they are not truly hosting these themselves.
The fact that they offer 'credits' rather than explain how many tokens you get, and do not indicate which 'frontier' models they have available just tells me some of this is, frankly, BS.
The only frontier models they will be able to host themselves are the Chinese ones as well, Anthropic, OpenAI etc do not make theirs available to 3rd parties.
For example, Deepseek V4 Pro needs around 960GB VRAM to run at 4bit quant. Kimi K3, at Q1 still needs around 600GB of VRAM - https://unsloth.ai/docs/models/kimi-k3 and 1.56TB at Q8. These are frontier models that are available for 'self hosting'.
The start up costs, for this, are staggering. Which is why I suspect heavily they are not truly hosting these themselves.
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff