Story

Hey HN. I've spent 10 years doing IT work, mostly infrastructure and scripting, the stuff nobody writes blog posts about. No CS degree. This is my first startup and I have no idea if I'm doing it right. Here's the problem that bugged me: every AI agent setup I looked at just blasts everything through GPT-4 or whatever the biggest model is. That's insane for 80% of tasks. You don't need a $0.03/1k token model to parse a CSV. So I built two things. Rhelm is a web app for organizing AI work visually, boards and tasks, kind of like Linear but for agent workflows. goClaw is a companion mobile app (coming to App Store and Play Store) that lets you talk to your agents from your phone. Text or voice. "Hey, spin up the data cleaning job" and it figures out which model to route it to based on what the task actually needs. The routing is the part I care about most. Task comes in, system looks at complexity, picks the cheapest model that can handle it. Sometimes that's a local model, sometimes it's an API call, sometimes it's "this actually needs a human." I built the core loop in about 4 days. I've been breaking and fixing it since. I'm not going to pretend I have everything figured out. I definitely don't. But I'm mass rounding off the rough edges and I'd rather ship something and learn from people smarter than me than sit on it forever. https://rhelm.io (waitlist is open) What am I missing? What would make you actually use something like this?