Questions about Operator, answered

Operator is an AI agent built from the coding agent you already use, and it never writes the code itself. A small command line tool gives your agent the ability to launch other coding agents as workers, hand each one a task, and review the code and test results they return the way a lead engineer reviews a pull request. The tool makes no decisions of its own, so every call about what to build and whether it's good enough comes from your agent.

Most orchestrators launch and track several agents, but leave it to you to split the goal into tasks and decide what runs in what order.

Operator splits the goal itself, or follows a plan you've already drafted, and reviews what each worker returns before deciding the next step. Weak work goes back with clear direction, and an approach that isn't working gets replaced, instead of everything running down a task list fixed at the start.

When you run them yourself, you split the work, write each prompt, check each result and decide what comes next. Operator takes over that job. It breaks the goal down, hands out the tasks, reviews what comes back, and checks in with you only where you've asked it to.

Operator is for developers who already work with a CLI coding agent and want it to take on larger goals. Instead of you breaking a project into prompts and supervising each one, Operator does the breaking down, the handing out and the reviewing.

Yes. Hand Operator a goal, your specs and rules, or a plan you've already drafted. It sticks to what you've decided and works out the rest as results come in.

Operator stops only at the checkpoints you set, and runs the job on its own between them. When one of its workers stops to ask a question or request permission, it asks Operator, not you.

Yes. Operator runs in your own terminal, in the session you started, so you can give it new direction at any point.

No. Every Operator session works in its own worktree, which is a separate copy of your repository on its own branch. Operator creates both when the session starts, so the copy you're working in is never touched.

Everything Operator and its workers produce lands on that session's branch, ready for you to review, merge or discard.

Operator and its workers run locally, on your own machine, in your own repository. Your code goes to the same coding agents, under the same accounts, as when you use them directly.

Operator is free to use, and it needs no new subscription or account. It has no model of its own: Operator runs on the coding agent and plan you already pay for, and its workers run on your existing accounts.

Operator works with a growing list of CLI coding agents. During setup you choose which agent your workers use and which model they run, and it can be the same agent Operator runs on or a different one. See which ones are supported today in the docs.

You need a CLI coding agent you already use and a git repository. From there it takes three steps:

• Install the ordi CLI.
• Run ordi init once in your repository to choose your agents and models.
• Run ordi oca to start Operator on a new branch of its own.

Then give it a goal. Operator plans the work, hands it to workers and reviews what comes back, while your own working copy stays untouched until you choose to merge.