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

Configuration¶

This page shows where Open Interpreter's settings live, how to choose a model and give it your key safely, and how to change the defaults.

Where things live¶

Path	What it is
`/usr/local/bin/interpreter`	The `interpreter` command. It runs Open Interpreter from `/opt/open-interpreter/venv` and refuses auto-run unless you allow it.
`~/.config/open-interpreter/profiles/default.yaml`	The `ubuntu` user's profile. Open Interpreter reads it when it starts.
`/var/lib/open-interpreter/profiles/default.yaml`	A reference copy of the image's profile.
`/var/lib/open-interpreter/workspace`	A working directory, owned by `ubuntu`, mode 0750.
`/opt/sushi/etc/open-interpreter/sushi-profile.yaml`	The template the profiles are created from.
`/usr/share/doc/open-interpreter`	License notices and corresponding source.

First boot creates each profile only if it does not exist yet, so your edits are kept across reboots.

The default profile¶

llm:
  model: "gpt-4o"
  temperature: 0

computer:
  import_computer_api: true

auto_run: false
safe_mode: "off"
offline: false
disable_telemetry: true
verbose: false

auto_run: false makes Open Interpreter ask before it runs each code block.
disable_telemetry: true turns telemetry off.
safe_mode: "off": Open Interpreter's optional code-scanning safe mode is not set up on this image. Your confirmation is the safety check, so read each code block before you approve it.

Edit the profile as the ubuntu user:

nano ~/.config/open-interpreter/profiles/default.yaml

Leave the version line at the end of the file unchanged.

Choose a model¶

Set llm.model to the model you want, in the LiteLLM model-name format that Open Interpreter uses. Then export that provider's key in your shell, as in Getting started.

To use an OpenAI-compatible endpoint, also set llm.api_base:

llm:
  model: "<model-name>"
  temperature: 0
  api_base: "https://<your-endpoint>/v1"

See the Open Interpreter documentation for the model names each provider uses.

Provider keys¶

Keep keys in environment variables, set in a private shell for the session. Never put them in EC2 user data or logs.

Do not add an api_key: line to the profile. First boot checks both profiles for a key on every boot. If it finds one, sushi-open-interpreter-firstboot fails, and the walkthrough refuses to run.

Allow auto-run for one session¶

The interpreter command refuses -y and --auto_run, so code always waits for your approval. To let Open Interpreter run code without asking in one session only, set SUSHI_OI_ALLOW_AUTO_RUN=1 for that command:

SUSHI_OI_ALLOW_AUTO_RUN=1 interpreter -y

Auto-run executes model output directly

With auto-run, any code the model writes runs as the ubuntu user, which can use sudo. Use it only on data and instances you can afford to lose, and never set the variable in ~/.bashrc.

Your data¶

Put the files you want to work on in /var/lib/open-interpreter/workspace and start interpreter from there. The code Open Interpreter runs can reach everything the ubuntu user can reach on the instance, plus anything the instance role allows, so keep unrelated data and permissions off it.

Last updated October 13, 2026 · Suggest an improvement