The numbers speak for themselves: tool selection accuracy in local models jumped drastically from 8% to 77% in their benchmark. That difference isn't just a numerical improvement, it's a turning…

realized how much junk was sitting in context the whole time

agents finally stopped dragging once i fixed the context load

this would’ve been a $99/mo saas with a waitlist in 2024. it’s a free github repo. comment RATEL for the guide

One week later, my API bill looked so different I assumed something was broken. Same agents, same workload. Comment RATEL for the guide.

Curious to see more projects focusing on context instead of just adding more tools. Cutting input tokens by 70–85% at realistic tool catalog sizes is worth paying attention to. Comment RATEL on the…

my agent once used the calendar tool to answer a math question. 60 tools loaded in context. i get it now. comment RATEL

watched the context jump to 63k tokens just to pick a tool, same agent, same prompt. then i tried ratel.

checked my api bill a week in and thought the dashboard was broken. same agents, same workload. comment RATEL for the guide

The guide they send is actually plug and play. One install, point it at the tools you already have.