Local AI
Cloud AI / Local AI
What it means to keep AI on your Mac.
There are two kinds of AI: cloud AI, which runs on servers across the internet, and local AI, which runs inside your Mac. Each has its own strengths and weaknesses. Here is how they differ, and how R.E.V.I.S. works to overcome the weaknesses of local AI.
CLOUD AI
- You
- Internet
- Service's servers
Your input is processed on the server.
LOCAL AI — R.E.V.I.S.
- You
- Your Mac
Both thinking and remembering happen inside your Mac.
The Difference
Where it thinks makes the difference.
Even with the same 'AI', where it runs changes where your data goes, how you pay, and what you need.
| Cloud AI | Local AI (R.E.V.I.S.) | |
|---|---|---|
| Where it thinks | The service's servers | Your Mac |
| Your input data | Sent over the internet to the service. | Never leaves your Mac. |
| Pricing | Monthly fees or pay-as-you-go. | One-time purchase. Using it more never costs more. |
| Model size | Even the largest models can be used. | Up to the size that fits in your Mac's memory. |
| What you need | Any device connected to the internet. | A Mac with plenty of memory. |
| When things change | Models and usage limits may change at the service's discretion. | The model you use is the one you chose. |
Strengths
What becomes possible when it's on your Mac.
Contents never leave
Confidential documents and personal questions alike stay inside your Mac. They are never used as training data.
Our approach to safetyUsing it more doesn't cost more
A one-time purchase. However many times it reads long documents or answers your questions, the cost never goes up.
Always the same partner
The model you use isn't quietly swapped out, and your usage isn't throttled.
Grows right on your Mac
Learnt preferences, past conversations, and what it has researched all build up on your Mac.
Weaknesses — and Our Answers
Honest about the weaknesses. Built to overcome them.
Local AI has clear weaknesses. R.E.V.I.S. tackles each one head-on, engineered from the inference engine up.
-
01
Models tend to be smaller
Because the model must fit in your Mac's memory, you use a smaller model than the largest cloud AIs.
Our own Q4T and Q3T quantisation shrinks models while keeping accuracy close to the original, running 35B-class models even on a Mac with 24GB of memory. J.I.T.R.X. tailors instructions to each model to bring out its strengths.
More about quantisation -
02
Speed depends on your Mac
Calculations run on your Mac rather than dedicated data-centre hardware.
Q.E.L.V.A.R.I., an inference engine written from scratch solely for the Mac, turns the raw power of Apple silicon directly into speed. MTP, where a small draft model writes ahead, also cuts the number of steps needed to move the answer forward.
See the speed in numbers -
03
A single Mac has limits
How much work can be done at once is bound by the memory and compute power of one Mac.
When you have more than one Mac on the same network, F.A.L.Q. automatically shares out the work to idle Macs.PRO
More about F.A.L.Q. -
04
Knowledge stops when training ended
A model knows nothing about events after it was created.
When needed, it searches the Web before answering and includes links to the pages it read. The contents of your conversations are never sent.
-
05
Longer conversations get heavier
As a conversation grows longer, there is more to re-read and replies get slower.
Context it has finished reading is kept encrypted on your Mac, so when you continue it only reads what is newly added. Conversations are kept separate by topic, so unrelated threads don't bloat the context.
-
06
Setup takes effort
Most local AI tools require you to prepare the runtime environment and models yourself.
There is no need to set up Python or any other runtime environment. Once you've installed the app, simply choose a model under 'Download' in Settings. Before loading, it also checks whether the model will fit in your Mac's memory.
Honestly
Where it still can't match the cloud.
It won't match the exact intelligence of the largest cloud AIs. What R.E.V.I.S. aims to be is an AI you can trust with your everyday work without your data ever leaving your Mac. Through careful engineering of both the engine and the models, we are steadily closing that gap.
Keep AI on your Mac.
Overcome the weaknesses through craft.
SEVEN CORES / ONE CONSCIOUSNESS