Skip to content

Getting started

Launch the image, run the local browser walkthrough, and optionally point the launcher at a model you choose.

You need: an AWS account that can subscribe and launch EC2. Session Manager needs an instance profile and the Session Manager plugin. SSH is an alternative when you restrict it to an administrative address. No model is required for the included fixture.

1. Subscribe

Open the Browser Use Browser Agent listing, choose View purchase options, review the terms, and subscribe.

2. Launch the instance

  1. Choose t3.medium or a larger listed type, with at least 4 GiB of memory.
  2. Use an encrypted gp3 root volume of at least 24 GiB.
  3. Require IMDSv2.
  4. For Session Manager, attach an instance profile that can use it. No application ingress rule is required.

3. Connect

Wait for cloud initialization, then run the walkthrough as ubuntu:

sudo cloud-init status --wait
sudo -iu ubuntu bash /opt/sushi/docs/demo.sh

The demo uses a local model fixture, the real Browser Use agent, and Chromium. It opens a synthetic page, submits a local form, and rereads three records whose values are 5, 17, and 20, totaling 42. A successful run prints SUSHI_CUSTOMER_DEMO_PASSED and the result directory. It does not buy anything and it does not submit to an external site.

Runs stay under /home/ubuntu/browser-use-results.

4. Add a model

Use an OpenAI-compatible chat-completions provider and a model that supports JSON schema structured output. Compatibility depends on that provider. Switch to ubuntu and run this in an unrecorded terminal:

sushi-browser-use configure

Enter the base URL, the exact model id, and the key. The key prompt is hidden, and setup does not send a model request. The command refuses to overwrite an existing settings file.

sushi-browser-use demo --output "$HOME/browser-use-results/provider-$(date -u +%Y%m%dT%H%M%SZ)"

That task sends the synthetic page and the instructions to your provider. Keep keys out of command arguments, task text, shell history, and recordings.

5. Reread a finished run

This does not call a provider and does not launch Chromium. Use the directory the walkthrough printed:

python3 -I -B /opt/browser-use/readback.py /absolute/path/to/the/completed/run

Last updated November 11, 2026 · Suggest an improvement