Developer documentation
Optimize your model for any hardware.
yasp takes a PyTorch model and generates optimized, hardware-specific inference artifacts. Agents do the compiler and kernel work, so you ship in days instead of months. No rewrites. No vendor lock-in.
Two products, one platform
Compile and deploy
From research code to a production runtime
One guided pipeline replaces weeks of environments, version management, conversion and unsupported operators. Compile, profile, optimize and deploy on your own hardware.
Start with Kronos →
Kernel generation
Peak throughput without hand-tuned kernels
Agents iterate on inference graphs and kernels deeper and longer than a human can sustain, for AMD, Nvidia and AWS Trainium targets.
Start with Gaia →
The command line
yasp-agent is the platform's command-line companion. It ships two console commands from a single package.
yasp
An interactive AI shell that wraps Claude Code with bundled yasp skills, hooks and permissions. Describe what you want and let the agent drive the compile flow.
yasp-toolkit
The scriptable surface: explicit commands for every yasp service. Built for automation, CI pipelines and users who prefer to drive each step themselves.