Skip to main content
Cadenza CLI is a command-line platform for physical-AI developers. You scaffold a robotics project, describe a multi-phase mission as JSON, run it in simulation, and turn the logged rollouts into fine-tuning data for a vision-language-action (VLA) policy, all from one binary called cadenza. It pairs with the Cadenza SDK (cadenza-lab), the Python library that provides the simulator, robots, action library, and inference stack the CLI drives under the hood.

​
What you can do

Scaffold projects

mkdir / env init generate a project with an env.json mission spec.

Run missions

env run executes a phase-aware mission, logs every tick, and scores it.

Build actions

env action creates reusable robot actions, synced to your account.

Fine-tune

env finetune / env lora turn rollouts into VLA training data and adapters.

​
How it runs

The cadenza binary works two ways:
  • Interactive shell. Run cadenza with no arguments to get a persistent prompt with tab-completion. Press Ctrl+C to exit.
  • Scriptable. Pass a command as arguments (cadenza env run my-project --headless) to execute it once and exit. This is what you use in scripts and CI.
Every example in these docs works in either mode. Inside the shell, drop the leading cadenza.
Simulation and fine-tuning features depend on optional extras (gym, lora). See Configuration.

​
Next steps

Install

Quickstart

Sign in