> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cadenzalabs.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> The Cadenza SDK: the Python library that powers Cadenza.

The **Cadenza SDK** is the Python library `cadenza-lab`: the MuJoCo-based
simulator, robots, action library, and inference stack that the
[CLI](/cli/overview) drives under the hood. Use it directly when you want to:

* Simulate robots — quadruped (Go1), humanoid (G1), and a 6-axis arm — and build
  scenes programmatically.
* Roll out policies through a Gym-style interface.
* Plug your own vision-language-action (VLA) / world model into the inference
  stack.
* Coordinate several robots on one goal over MCP.
* Drive a CLI project's mission with your own model.

## Install

```bash theme={null}
pip install cadenza-lab
```

Or get it transitively via a CLI source install with the `gym` extra:

```bash theme={null}
pip install -e ".[gym]"     # from the cadenza-cli repo
```

## Import

The PyPI distribution is `cadenza-lab`. The importable module is `cadenza`. A
thin alias lets you import it under either name:

```python theme={null}
import cadenza_lab as cadenza   # recommended
# or:  import cadenza
```

## The API at a glance

| Area                         | Entry points                                                                             | Guide                                                   |
| ---------------------------- | ---------------------------------------------------------------------------------------- | ------------------------------------------------------- |
| **Quick rollouts**           | `cadenza.run(...)`, `cadenza.view(...)`                                                  | [Quickstart](/sdk/quickstart)                           |
| **Robots**                   | `cadenza.go1(...)`, `cadenza.g1(...)`, `cadenza.arm(...)`, `Step` / `ArmAction`          | [Robots](/sdk/robots)                                   |
| **Actions**                  | `list_actions`, `get_action`, `get_library`, `ActionLibrary`, `ActionSpec`, `ActionCall` | [Actions](/sdk/actions)                                 |
| **Multi-robot coordination** | `cadenza.connect(...)`, `term.coordinate(...)`, `robot.comm`                             | [Coordination](/sdk/coordination)                       |
| **Natural language**         | `cadenza.parser.CommandParser`                                                           | [Actions](/sdk/actions#natural-language-parsing)        |
| **Simulation**               | `Sim`, `Scene`, `Box`, `Sphere`, `Slope`                                                 | [Simulation & scenes](/sdk/simulation-and-scenes)       |
| **Gym interface**            | `GymAdapter`, `Observation`                                                              | [Gym adapter](/sdk/gym-adapter)                         |
| **Inference orchestration**  | `Sequential`, `ChainOfThought`, `InferenceOrchestrator`                                  | [Inference orchestration](/sdk/inference-orchestration) |
| **World-model stack**        | `cadenza.stack.run(...)`, `register_world_model`, `WorldModelAdapter`, `Modality`        | [Inference stack](/sdk/inference-stack)                 |
| **Deploy**                   | `go1.deploy(...)`, `go1.deploy_ssh(...)`, `go1.deploy_ssh_bridge(...)`                   | [Deploy](/sdk/deploy)                                   |

## How the pieces fit

Everything in the SDK is layered on one physics core. You can enter at any layer:
fire a one-liner, script `Step`s, step a Gym loop, or hand the whole loop to a
world model.

```mermaid theme={null}
flowchart TD
    NL["Natural language<br/>'walk forward then sit'"] --> P[CommandParser]
    P --> AC[ActionCall]
    Steps["go1.stand(), go1.jump()<br/>(Step descriptors)"] --> AC
    WM["World model / VLA<br/>(WorldModelAdapter)"] -->|ProposedAction| BLD[Action builder] --> AC
    AC --> LIB[(Action library<br/>gaits + phases)]
    LIB --> SIM[Sim · MuJoCo physics]
    GYM[GymAdapter] --> SIM
    SIM -->|Observation| GYM
    GYM -->|obs| WM
    SIM --> DEPLOY[Deploy drivers<br/>DDS · SSH · bridge]
    DEPLOY --> HW[Physical Go1 / G1]
```

<Card title="Quickstart" icon="bolt" href="/sdk/quickstart">
  Run a robot and render a scene in a few lines.
</Card>