Your Mac is already an AI computer

Apple is often described as the company that missed the AI boom. We think it quietly built something more useful: a computer that can do most everyday AI work by itself.

That idea is what Rooster is built on. Many everyday AI jobs, like transcribing a recording or cutting out the ums, don’t need a data centre. Doing them on a Mac you own, instead of paying by the minute for someone else’s servers, is cheaper, faster and better for the environment. Cloud-based AI is for the jobs a Mac can’t handle.

The hardware is already on your desk

Every Apple silicon Mac since the M1 has a Neural Engine - a part of the chip built for machine learning - next to its CPU and graphics processor. All three share the same memory, so a model and the video it is working on sit in one place instead of being copied back and forth. Which part does the work depends on the model and the app; Apple’s machine-learning overview explains the options.

You paid for this when you bought the Mac, and most of the time it sits idle. Software that uses it isn’t asking you to buy anything new.

Most AI in video editing is small work

A lot of AI talk is about huge models that write, reason and plan. Editing a talking-head video mostly needs something narrower:

• turning speech into words with accurate timings
• finding the ums, the uhs and the long pauses
• separating a voice from background noise
• separating a person from the background behind them

Each of these is a specific job, and a model built for that job runs well on a Mac. Once the transcript has timings, deleting a sentence is ordinary editing. Rooster already knows which part of the recording to cut, so pressing Delete doesn’t send anything anywhere.

Owning beats renting

When the work happens on a server, someone pays for that server every minute it runs. That cost comes back to you, which is why cloud editors sell media minutes, AI credits and top-ups. Record more, and the meter moves.

When the work happens on your Mac, another hour of transcription costs us nothing, so Rooster doesn’t meter it. Record another take, try a different cleanup setting or re-cut the whole thing without checking a balance.

It’s faster too. Nothing has to upload, so transcription starts as soon as you stop recording, and it still works on a plane.

Better for the environment

Doing the work on a computer you already own means no extra hardware has to be built and kept running for it. Your footage isn’t uploaded, copied and stored in a data centre. And a small model made for one job does that job without the overhead of a giant general-purpose one.

We haven’t measured the difference, so we won’t put a number on it. But using no more computing than a job needs seems like the right default.

The cloud still has a job

We use cloud AI every day ourselves. It’s the right tool for work a Mac can’t do well, such as research, long writing, or generating an image or a clip from nothing. Rooster connects to a cloud service when you want one - your own transcription provider, with your own key - and you pay that provider directly.

What we don’t accept is that routine editing should go through a server by default. Apple has already put capable AI hardware in the computers people make videos on. So why keep paying by the minute for work your Mac can do?