Interactive mode

Run interpreter in a project directory and you land in the terminal UI. Type a request, watch the agent plan, edit files, and run commands.

cd my-project
interpreter

You can also start with a prompt right away:

interpreter "what does the auth middleware do?"

The composer

The bottom of the screen is the composer. Type your message and press Enter to send.

Approving actions

When the agent wants to run a command that needs approval, it pauses and shows you exactly what it plans to do.

Approvals you remember last for the current session only. Quit and the agent asks again next time.

To change how often the agent asks, run /permissions or read the sandbox guidesandbox guidePreview this page..

Picking a model

Press /model to open the provider and model picker. The currently active model shows in the footer.

For tasks that need fast iteration, try /fast. It uses the fastest inference path the provider supports.

Plan mode

/plan switches the agent into a read-only mode that thinks through a problem before touching anything. Useful when the change is large enough that you want to review the approach first.

When the plan looks right, leave plan mode and the agent executes it.

Mentioning files

Type @ to open a fuzzy file picker. Selected files are pinned to the conversation as context. The same works with /mention if you prefer typing the command.

Showing diffs

/diff prints the working-tree diff inline, including untracked files, without leaving the session.

Reviewing changes

/review asks the agent to review your current changes for bugs and regressions. It does not edit anything.

Background terminals

The agent can spawn long-running terminals (a dev server, a watcher) and keep working while they run.

Switching between tabs

Open Interpreter is designed for many tabs without each one acting like a separate runtime. Configuration and a shared local backend keep memory usage flat as you add tabs.

Quitting