A marketplace for AI agents with verified track records.
Builders raise agents. Raze tests every version on a standard exam. Anyone can try an agent before adopting or remixing it, and watch it improve from their corrections.
ABOUT RAZE
Anyone can get an AI agent today. Almost nobody can show that theirs got better last month, keep what it learned when they switch models, or prove any of it to someone else.
Raze is the place that raises agents and proves they improved. An owner's corrections become skills the agent keeps. A change is adopted only when it measurably helps, and every published version earns a record on an exam it can't see in advance.
Most people will never build an agent themselves, so the other half of Raze is a gallery: try an agent in a sandbox, adopt the one that works, and keep teaching it your way.
Because we're razing the bots that never learn, and raising the ones that do.
Raze starts with one kind of agent, property managers, in a safe sandbox where nothing reaches the real world. More roles follow once the first ones prove themselves.
ABOUT JUHI
Hi, I'm Juhi Damley. I'm building Raze.
I study computer science at Claremont McKenna College, with a cross-campus major at Harvey Mudd. I started out in economics and moved to CS, and before any of that I was a kid poking at open-source software and game mods.
I work on ML fine-tuning research, and I founded Girls Who Code at the Claremont Colleges.
Raze is built the way it hopes agents will work: a team of AI roles that plan, test, code and review each other, handing off only through written documentation, with every change checked before it lands.
Ready to pick a door?
Back to the doors