Operator — autonomous goal runs

What it makes

Operator takes a goal and works at it on its own. It writes a task queue first, runs one task at a time, reviews what came back after every result and re-plans when the result changes the picture. At the end it hands you one written deliverable rather than a transcript to sift through.

For research, comparisons and reports — work that takes twenty steps you would otherwise run yourself, one browser tab at a time.

How a session works

1. Write the goal in a sentence or two and enable the tools this run is allowed to use.
2. Read the plan it proposes before anything runs, and let it start when the plan looks right.
3. Watch the queue work through, then collect the single deliverable it writes at the end and share it by link.

What it costs

Every phase of a run is a model turn, so planning costs credits too — a small one. Each task the queue executes spends more, so a long goal costs more than a short one, and the plan tells you how long it will be before anything runs.

Generation spends credits: paid plans include a monthly allowance, the Free plan starts with a one-time welcome grant, and you can top up at any time.

Included from the Pro plan.

Questions

What can Operator actually do?

It researches, gathers, compares and writes. The tools it may use are the ones you enable for that run, so its reach is exactly as wide as you decided before it started.

Does it run without asking me?

It runs the plan without asking at every step, which is the point, but you see the plan before anything happens and you can stop the run at any time.

What do I get at the end?

One written deliverable — the answer, the report or the comparison you asked for — with a link you can share, rather than a log of everything it tried.

Which tools are on by default?

None. Tools are enabled per run, never by default, so a run cannot reach for something you did not give it for that goal.