Operator

Let Cerevisor check in on a schedule while you're away, make one small bounded move at a time, and wait for your review before anything sticks.

Operator checks in on your workspace every so often while Cerevisor is open, makes one small, bounded move, and then goes back to waiting. It's meant for the moments you're not at your desk — a lunch break, an overnight stretch, a weekend — when there's real work it could be quietly pushing forward instead of sitting idle.

Plan requirement: available on trial and paid plans.

What it does while you're away

Every check-in, Operator looks at what's going on in your workspace and does exactly one of three things:

  • Moves an objective forward. If you're already tracking something it can help with, it advances it by one small step — the same kind of step you'd approve if you were driving it yourself through chat.
  • Starts tracking something new. If the evidence clearly points at something you're working toward that nothing is tracking yet, it can start tracking it. This uses the paid request that chooses the next move, but does not start a separate build or run. Operator states in plain language what finishing the thing would actually mean.
  • Makes itself more useful for next time. If there's nothing to move forward, it spends a little of its budget drafting something — a skill, a Power, a memory entry, a workflow sketch, or a short summary of your workspace — for you to look at later.

If it runs into a question it can't answer on its own, it stops and flags it for you rather than guessing. And if the last few real attempts in a row didn't go anywhere, it turns itself off automatically and tells you, rather than continuing to spend on something that isn't working — see When it turns itself off below.

Operator only checks in while Cerevisor itself is open. Closing the app pauses it completely. There is no separate service checking in while your computer is asleep or the app is closed.

It spends real money

Operator uses the same connected AI provider account as everything else in Cerevisor — there's no separate meter. Before turning it on, you set two limits in Settings → Operator:

Setting What it controls
Daily budget Once today's recorded spend reaches this amount, Operator makes no more paid requests until the next day. You'll get a notification when that happens.
Per-check limit Bounds the work used to move an objective forward after Operator chooses it. The initial paid choice and self-improvement drafts sit outside this particular limit.
How often it checks From every 5 minutes up to every 12 hours. Thirty minutes by default.

Self-improvement drafting has a smaller daily allowance: a quarter of the daily budget by default. Once its recorded spend reaches that allowance, Operator starts no more drafts that day.

A check that starts while budget remains is allowed to finish. Its final cost can take the recorded total past the daily amount or the drafting allowance; those walls prevent the next paid check or draft. If you lower the daily budget below what you've already spent, the new limit blocks the next paid check. Cerevisor also lowers any per-check setting that is higher than the daily budget.

Reach: how much it's allowed to touch

Reach is the safety setting for a self-directed build and run. It has two positions, and you change it in Settings, not on the Operator panel itself:

  • Sandboxed (default). The work is confined to a separate folder. Internet access, outside services, package installation, source-control changes, and command-line programs are blocked.
  • Full access. The work may use your real project folders and the permissions already attached to the work it creates, including access to outside services. This is a meaningfully bigger trust decision, so switching to it requires an explicit confirmation in Settings. If that confirmation was never recorded, Cerevisor treats the setting as Sandboxed. Full access cannot turn itself on silently.

The self-directed build-and-run action is implemented but is not currently offered to Operator, because its changes cannot yet be safely applied to the workspace you have open. Until that is released, Reach does not change the five drafting actions Operator can choose. Those drafts always go to the review queue. The setting is visible now so its safer default and confirmation are already in place before that action is enabled.

Work on an objective you're already tracking always runs with that project's own normal permissions — the same access it would have if you ran it yourself — regardless of the reach setting, since it isn't self-improvement drafting.

Everything it drafts waits for your review

Nothing Operator drafts — a skill, a Power, a memory entry, a workflow sketch, a workspace summary — takes effect the moment it's created. Every draft lands in a review queue where you can read it, see what it's based on, and choose Accept or Discard. If a draft is based on evidence Cerevisor couldn't fully verify, the queue flags that clearly so you can weigh it before deciding. Nothing is applied automatically, ever.

When it turns itself off

After three unsuccessful attempts in a row — failures or times it has to stop and ask you for information — Operator turns its own autonomy off and tells you. A budget pause or a skipped check because something else is running does not count and does not clear that streak. You'll see why in the journal, and you can turn it back on once you're ready.

Reviewing what it did

Open the Operator panel to see:

  • The journal — a plain-language, most-recent-first log of every check-in: what it did, and what it cost. This is the place to build trust in what Operator is doing while you're not watching.
  • Objectives — what it's currently tracking, with status and any question it's waiting on you to answer.
  • The review queue — everything it's drafted, waiting for Accept or Discard.
  • Stop — turns autonomy off immediately.

Checking in from your phone

If you've connected the Cerevisor Companion app, your phone shows a lighter version of the same picture: the journal feed, any question Operator is waiting on you to answer, today's spend, and a Stop button. Reviewing and accepting drafts stays a desktop-only step for now.

Turning it off (the kill switch)

You can stop Operator from either the desktop panel's Stop button or your phone. Either one turns autonomy off right away — no more check-ins will start.

One thing Stop can't do: cancel a check that's already in progress at the moment you press it. If one was mid-way through when you stopped it, it finishes on its own, and you may see a small charge from that last check even though Operator is now off. Everything after that stays off until you turn it back on.

Cost expectations

The cost of a check depends on your provider, model, and the work chosen. Choosing what to do is itself a paid request; advancing work or creating a draft can add more paid requests. A check blocked before that choice — for example because the daily budget is exhausted or another run is active — makes no provider request and costs nothing. Set the daily budget to an amount you're comfortable letting Cerevisor spend on its own in one day. You can raise or lower it later.

Back to docs