Skip to content
Open Interpreter AI Agents · Open Interpreter 0.4.2+sushi.2 Launch on AWS

Getting started

This page takes you from the AWS Marketplace listing to your first Open Interpreter task: subscribe, launch, connect, check the installation and run it with your own model provider. It takes about ten minutes.

You need: an AWS account with permission to subscribe to AWS Marketplace products and launch EC2 instances, and an API key for a model provider.

1. Subscribe

Open the Open Interpreter listing, review the pricing and terms, and subscribe.

2. Launch the instance

Launch from the listing with these settings:

  • Instance type: t3.medium.
  • Storage: at least 16 GiB, encrypted gp3.
  • Metadata: IMDSv2.
  • Security group: no inbound rules are needed. Allow outbound HTTPS.
  • IAM instance profile: attach a role with the AWS managed policy AmazonSSMManagedInstanceCore so you can connect with Session Manager. Open Interpreter itself needs no AWS permissions.

3. Connect

Connect with Session Manager, from the EC2 console (Connect > Session Manager) or with the AWS CLI:

aws ssm start-session --target <instance-id>

Wait for first boot to finish:

cloud-init status --wait

4. Run the walkthrough

sudo -iu ubuntu bash /opt/sushi/docs/demo.sh

The walkthrough checks that first boot succeeded, prints the installed version, checks that the profile has auto-run off, telemetry disabled and no API key, runs a small Python calculation locally, confirms the interpreter command refuses -y, starts the interactive prompt once in offline mode, and then runs the self-test. No model is called.

Each step prints STEP, COMMAND and RESULT lines, and the script ends with SUSHI_CUSTOMER_DEMO_PASSED.

5. Set your provider key

Switch to the ubuntu user:

sudo -iu ubuntu

The profile uses the model gpt-4o by default. For OpenAI, set your key in this shell only. With read -s, the key does not appear on screen or in your shell history:

read -rs OPENAI_API_KEY && export OPENAI_API_KEY

For another provider, change the model in the profile and export that provider's key instead. See Choose a model.

Keep your key out of the profile

Do not write api_key: into the profile. The image treats a key in that file as a mistake: the first-boot service fails on the next boot and the walkthrough refuses to run. Use an environment variable instead.

6. Run your first task

Go to the workspace and start Open Interpreter:

cd /var/lib/open-interpreter/workspace
interpreter

Describe what you want, for example "create a CSV with the numbers 1 to 10 and their squares, then print the sum of the squares". Open Interpreter explains its plan and shows the code it wants to run. Read it, then approve or refuse.

To leave, press Ctrl+D. Open Interpreter may ask whether you want to send feedback; answer no to share nothing.

Next steps

Choose a model, use an OpenAI-compatible endpoint and change the defaults in Configuration.

Last updated October 13, 2026 · Suggest an improvement