AI instruction sets are software.
They should be readable, versioned, and improvable by the people who use them.
What we believe
Most AI products treat their prompts the way traditional software treated source code before open source: proprietary, hidden, and impossible to audit. We think that's the wrong model for tools that sit this close to how people think and work.
When you use a Kumo kit, you're not buying a black box. You're buying a well-structured starting point with a known design rationale — and you can read every line of it on GitHub.
What we build
Kumo sells Claude Project Kits: instruction packages for Claude scoped to specific roles, workflows, or life situations. A kit contains the system prompt, example interactions, and documentation explaining the design choices. You load it into a Claude Project and get a configured AI with a clear function.
We also take on a small number of web design and web app projects each year. Fixed price, milestone payments, two active projects at a time.
Why we publish everything
Publishing the instructions is a constraint we imposed on ourselves. It means every kit has to hold up to scrutiny. We can't ship something that only works under the right framing — it has to be legible, defensible, and genuinely useful to someone reading cold.
It also means you can evaluate what you're buying. Read the repo. If the instructions look thin, don't buy the kit. If they look solid, you'll know why the kit is priced the way it is.
We think this is how AI tooling should work.
What we're not
We're not a prompt-engineering consultancy. We're not a course. We're not a community with a monthly fee. We build specific tools for specific problems and publish what's inside them.