Canyon platform documentation
Canyon turns natural-language intent into running, data-connected applications, deployed inside your own infrastructure. These pages describe how the platform is assembled, how agents reason, how the tool layer reaches your systems, and how the whole stack deploys into your cloud.
What is Canyon
Canyon is an enterprise AI production platform. It combines a multi-agent code-generation pipeline, a governed semantic data layer, a W3C DTCG design-token system and a sandboxed preview runtime into a single control plane that deploys to your own cloud.
Canyon is designed for the enterprise reality: SSO you already run, data that cannot leave your perimeter, procurement that expects certifications, and architectures that must survive audit. The pages below describe how each layer of the platform meets that bar.
At a glance
Reference map
Every page below describes one subsystem of the running platform. Each opens with the concepts, lists what is live today, what is in alpha, and what you can request. The reference diagrams are the same ones the engineering team uses.
Suggested reading order
Start with System architecture for the high-level view, then follow the pipeline: Agent orchestration → MCP integration → Themes & tokens. After that, pick the branch that matches your concern: platform (auth, capabilities), data layer, or infrastructure.