Deployment

Language / 语言： English | 简体中文

This page is for people deploying the standalone application or maintaining its hosting.

Production (Cloudflare Workers)

Live site: https://lerobot.studio

Connect ioai-tech/lerobot-studio to Cloudflare Workers Builds so merges to main deploy without storing a Cloudflare API token in GitHub. Restrict production deploys to protected main; do not give fork PR workflows Cloudflare credentials.

Config: wrangler.jsonc

• Worker: lerobot-studio
• Assets: dist
• SPA fallback: not_found_handling = "single-page-application"
• Domain: lerobot.studio

Recommended Workers Builds settings:

Setting	Value
Build command	npm ci && npm run build
Deploy command	npx wrangler deploy (or Cloudflare default)
Root	repository root
SITE_URL / VITE_SITE_URL	https://lerobot.studio

Local dry-run:

SITE_URL=https://lerobot.studio npm run build
npx wrangler deploy

Docker

docker build -t lerobot-studio .
docker run --rm -p 8080:8080 lerobot-studio

Or use the root docker-compose.yml (maps host port 8080):

docker compose up --build

Optional build args:

• BASE_PATH — subdirectory deploy path (e.g. /lerobot/)
• SITE_URL — canonical site URL for SEO assets (production default: https://lerobot.studio)
• VITE_SAMPLE_DATASETS_MANIFEST_URL — full URL to sample-datasets.manifest.json (when unset, the welcome page hides the samples list)

Static hosting

SITE_URL=https://lerobot.studio npm run build

Serve dist with any static file server. Configure SPA fallback to index.html.

Documentation site (GitHub Pages)

The documentation is a separate VitePress build:

npm run docs:build

.github/workflows/docs-pages.yml builds and deploys docs/.vitepress/dist after documentation changes reach main. Its default DOCS_BASE is /<repository-name>/, so it works at the GitHub project Pages URL without assuming that a custom documentation domain exists.

The repository is configured to deploy its documentation through GitHub Actions. When maintaining that deployment:

1. Keep Settings → Pages set to GitHub Actions.
2. Verify https://ioai-tech.github.io/lerobot-studio/ after changes to the workflow or domain settings.
3. Add the workflow’s build and deploy jobs to the protected-branch policy if documentation deployment is required for merges.
4. If you add a custom domain, configure it in GitHub Pages, add the required DNS records, enable HTTPS, and verify the domain before publishing the URL.
5. For a root custom domain such as docs.lerobot.studio, change the workflow DOCS_BASE to /. Do not add a CNAME file or advertise that hostname until GitHub Pages and DNS both confirm it.

The workflow requires pages: write and id-token: write only in the deploy job. Pull requests build the same site through the read-only CI job but do not deploy it.

Sample datasets and CORS

Sample dataset archives are not shipped in this repository. Production or self-hosted builds may inject a remote manifest with:

• VITE_SAMPLE_DATASETS_MANIFEST_URL — full URL to the manifest JSON.

Relative archiveFile / coverImageFile paths resolve against the manifest directory, or against optional manifest.baseUrl when present. When the env var is unset, the welcome page hides the samples list. Generate a manifest on a deployment machine that holds authorized assets:

node scripts/generate-sample-manifest.mjs \
  --dir ./storage \
  --baseUrl https://examples.example.com/samples/ \
  --out ./temp/sample-datasets.manifest.json

Remote archives must allow cross-origin GET. Scalable access to large archives also requires byte ranges. A typical object response includes:

• Access-Control-Allow-Origin: https://your-viewer.example (or * for a genuinely public, credential-free object);
• Accept-Ranges: bytes;
• Content-Length;
• a 206 Partial Content response with a valid Content-Range when the client sends Range: bytes=...; and
• Access-Control-Expose-Headers: Accept-Ranges, Content-Length, Content-Range when those headers must be read by browser code.

If a preflight occurs, allow GET, OPTIONS, and the Range request header. Do not use wildcard origins with credentialed requests. A server that ignores Range may trigger a full download and is not a supported large-dataset setup. See Compatibility for the complete network boundary.

Privacy

The npm library and open-source web shell do not include usage analytics. Deployers are responsible for documenting any logs, analytics, proxies, sample endpoints, authentication, and retention policies they add. See Privacy.