Interactive Mode

Run interpreter in a project directory to start the terminal UI:

cd my-project
interpreter

You can also pass the first prompt on the command line:

interpreter "find the auth middleware and explain how it works"

Composer

The composer is the prompt box at the bottom of the TUI.

Files and Images

Use @ to fuzzy-search files and add them as context. You can also attach images to the first prompt:

interpreter -i screenshot.png "explain what is wrong in this UI"
interpreter -i before.png,after.png "compare these states"

Approvals

When a command or tool needs approval, the TUI shows the request before it runs. The default posture is designed for day-to-day work in a trusted repository: workspace access is allowed, and actions outside the active policy ask first.

Change the active policy with:

/permissions

For details, see Sandbox & approvals and Permissions.

Models and Providers

Use /model to pick the provider, model, and reasoning effort. Open Interpreter supports OpenAI, Anthropic, local providers, and compatible custom providers from the generated model catalog.

Common one-off overrides:

interpreter -m gpt-5.1-codex "review this module"
interpreter --oss "try this with my local model"

Review and Planning

Use /plan when you want the agent to inspect and propose before it edits. Use /review when you want a code-review pass over current changes.

/plan
/review

Review mode is read-focused. It reports bugs, regressions, missing tests, and risky behavior before summaries.

Background Work

Long-running commands can stay alive in background terminals while the agent continues.

Session Controls

Open Interpreter keeps session state locally under ~/.openinterpreter/.