Skip to main content
RLDX-1 is a vision-language-action (VLA) model by RLWRLD. It reads three camera views, robot state, and a task description. It returns chunks of actions. It does not generate video. In this guide, you deploy RLDX-1 on Reactor using a prebuilt model image. Then you test the deployment with a Python client that sends synthetic camera frames and robot state. The complete example in the Reactor cookbook includes the model code, deployment configuration, and test client. This example enables guided Real-Time Chunking (RTC). The model uses three 256 × 256 RGB views and returns 16-step action chunks over the data channel.

​
Requirements

You need Git, a running Docker engine, and a Reactor account with deployment capacity in your selected region. Your laptop does not need an NVIDIA GPU. Your account needs deployment access. Contact us if your account does not have access. The cookbook code is Apache-2.0. The checkpoint uses the separate RLWRLD Model License. Review its use restrictions before you download or deploy the weights.

​
Deploy the model

​
1. Clone the cookbook

​
2. Install the CLI and sign in

Complete the browser sign-in with the account that will own the deployment. See Install the CLI for installation options and Authenticate for CLI sign-in details.

​
3. Pull the model image

Start Docker. Pull the published image:
Keep the image tag 1.0.1 when you publish. This image tag is separate from the model release. The CLI reads model.name and model.version from reactor.yaml.

​
4. Open the model workspace

The workspace points to the public Hugging Face checkpoint:
The CLI records the pinned reference. Reactor fetches the checkpoint when the model starts. The example reactor.yaml requests one instance in us-west by default. If you use it unchanged, your account needs capacity in us-west. To use another region, change region under deployment.instances in reactor.yaml before you deploy.

​
5. Publish, deploy, and check status

Run each command from models/rldx-1.
publish --source pushes the downloaded image into your account without rebuilding it. It associates the checkpoint with the release. deploy activates the release and applies the instance plan. Wait for the deployment to become ready. Startup includes fetching the checkpoint and loading the model. Keep the account-qualified model name that the CLI prints, such as your-account/rldx-1. Next, test your deployment with the matching Python client. For the deployment workflow, see Deploying models. For checkpoint options, see Load your weights. The CLI reference covers publish, deploy, and status.