Skillbooks is our bet on what AI is missing: a real knowledge layer. Not just smarter models, but a way for agents to load expertise, combine methods with source material, and become trustworthy in serious domains.
AI agents are making things up at scale. Not because the models are useless — because the layer beneath them is weak: stale training data, noisy web search, and brittle retrieval systems masquerading as knowledge.
We think the next era of AI depends on a better abstraction. Every skillbook is either a reference (what you know) or a guide (what you do). References carry source material. Guides carry method. Load them together and an agent can do something much more powerful than autocomplete: it can reason with real expertise.
That matters for trust. It matters for economics. And it matters because experts deserve a future where their knowledge becomes part of the AI stack without being stripped of credit, control, or upside.
The result we want: agents that can show their work, creators who get paid, and a world where expertise becomes composable infrastructure.
Valuable knowledge gets absorbed into training data. Authors see no upside. That's broken.
Without structured sources, AI makes things up. Confidently. At scale. In production.
Users can't trust answers. Authors can't monetize. Developers maintain fragile pipelines forever. We're done with that.
Reference + Guide, composed on demand. Domain expertise that arrives instantly and can be swapped the moment something better exists.
80% revenue share. No free tier. Authors earn from the first page load. Knowledge that compounds while you sleep.
We build on existing formats and ecosystems — Markdown, HTTP, plain GET requests. No new protocols to learn.
Every claim on this site is backed by working code and real examples. Agents are reading skillbooks right now.
The format spec is open. The CLI tools are open. We believe the best way to build trust is to build in public.