Skip to main content

Documentation Index

Fetch the complete documentation index at: https://libretto.sh/docs/llms.txt

Use this file to discover all available pages before exploring further.

Run Libretto sessions on Steel’s managed browser infrastructure.
Steel is a managed browser provider. Use it when you want Libretto’s local CLI and workflow runtime, but want Steel to host the browser session.

​
Configure Steel

Create a Steel API key, then add it to the repo root .env file. Libretto loads .env automatically before running CLI commands. 
# inside .env
STEEL_API_KEY=<steel-api-key>
STEEL_BASE_URL is only needed if you use a custom Steel API endpoint. STEEL_CONNECT_URL is only needed if Steel gives you a custom CDP connection endpoint. Then set Steel as the workspace’s default browser provider in .libretto/config.json: 
{
  "version": 1,
  "provider": "steel"
}
You can also use --provider steel on a single command instead of changing the config file.

​
Open a session

Open a page with the configured Steel provider: 
npx libretto open https://example.com --session steel-demo
If you did not set provider in .libretto/config.json, pass it for this command instead: 
npx libretto open https://example.com --provider steel --session steel-demo
After the session is open, use the normal Libretto commands: 
npx libretto snapshot --session steel-demo
npx libretto exec --session steel-demo "await page.title()"
npx libretto close --session steel-demo

​
Run a workflow

Run workflows normally when provider is set in .libretto/config.json: 
npx libretto run ./workflow.ts --headless
Or pass the provider for a single run: 
npx libretto run ./workflow.ts --provider steel --headless
The workflow code does not need to change. Libretto creates the Steel browser session, connects to it over CDP, runs the workflow, and releases the provider session when the run finishes.

​
When to use Steel

Choose Steel when you want a managed browser runtime and already use Steel for browser sessions. If you need to own the container runtime, scheduler, network, and secrets layer yourself, use GCP or AWS instead.