Topic guide

Deployments

Use deployments for more control over how your policies run. The tested package-first path starts in the open-source runrobot repo, then publishes into RunRobot Cloud for hosted deployment control.

Open runrobot GitHub repo Open deployments console

Step 1

Package the policy first

The deployment boundary is the package contract. Train with any stack you want, export into manifests, and validate that package before you ever create a hosted deployment.

php-parkour/
  policy.yaml
  robot.yaml
  scene.yaml
  validator.mjs

Create policy.yaml, robot.yaml, scene.yaml, and a validator.
Validate the package locally in open-source runrobot.
Open the same package in the browser and verify it behaves correctly.
Publish the exact package to RunRobot Cloud.
Create a deployment only after the published run already works.

Step 2

Open a published policy run

The shortest developer path is still URL-based. Build the canonical browser run URL for the published policy, then open it directly.

import Runrobot from "runrobot";

const client = new Runrobot({
  auth: process.env.RUNROBOT_API_TOKEN ?? "YOUR_RUNROBOT_API_TOKEN",
});

const url = client.buildRunUrl({
  owner: "amazon",
  policySlug: "php-parkour",
  robotRef: "unitree/G1",
  sceneRef: "chasebrignac/parkour",
});

console.log(url);
// https://runrobot.sh/amazon/php-parkour/unitree/G1/chasebrignac/parkour

What are deployments?

Deployments give you production-grade control over how you run a policy. They provide stable, dedicated policy endpoints on top of packaged and published runs.

Hardware flexibility

Choose the runtime family you want for a policy. Browser MuJoCo and server MuJoCo stay close to the same package contract, while native robot lowcmd is the path to hardware execution.

Intelligent scaling

Set minimum and maximum instances, scale to zero when traffic is quiet, or keep policies warm so they avoid cold starts.

Zero-downtime deployments

Deployments let you roll a policy forward without changing client code. The stable endpoint stays the same while the targeted policy package changes underneath it.

Production monitoring

Monitor whether policy instances are starting, idle, or processing. Rollout logs and deployment records stay attached to the same policy lifecycle in the RunRobot console.

Enterprise security

The hosted control plane can expose dedicated policy endpoints, audit-friendly deployment records, and on-prem-compatible deployment flows.

Autoscaling

Deployments scale up when demand arrives and scale back down when it leaves. Minimum and maximum instance settings let you control cost and responsiveness.

Tested example

PHP Parkour

The current tested example is the php-parkour package based on the open-source runrobot repo. It uses the `browser.mujoco.onnx` contract with the default G1 robot and parkour scene.

Open hosted php-parkour Back to docs