Why we design for 8 GB laptops

Most AI software quietly assumes you have a powerful computer. We don't. Our target is the laptop that, by our estimate, most people in India own: 8 GB of memory, a browser with too many tabs, and Excel or your accounting software open in the background.

The short answer: we design Pivra Desk for 8 GB computers because, by our estimate, most computers in use in India have 8 GB of memory or less, and AI that only works on expensive machines leaves most people out. Today that means memory checks before a model loads and smaller models to choose from; we will measure 8 GB Macs and Windows laptops and publish the numbers.

This is the third post in our series for India. The manifesto explains why we believe AI should run on the computer you already own. This post is about what that means when the computer is modest.

Why 8 GB#

We looked at public data on PC shipments and memory in India. No research firm publishes this split publicly, so we built our own estimate from what is published: shipment volumes, price bands and how long people keep their laptops.

Our estimate is that most computers in use in India today have 8 GB of memory or less, and that this will still be true for several more years. Laptops are bought carefully, kept for years and often passed on, which we think is sensible, not a problem to be fixed.

We hold this estimate with low to moderate confidence on the exact share, and more confidence on the direction. That is enough to make a decision: if we design for 16 GB, we design for a minority.

So 8 GB is not an edge case for us. It is the computer we design for.

What designing for 8 GB means, and what it doesn't#

We are careful with words here. "Designing for" is a commitment about how we build. It is not a claim that everything is already fast on every 8 GB machine.

What it means today:

It checks memory before it loads a model#

Before Pivra Desk loads a model on your computer, it checks how much memory is free right now and picks a context size that fits. If memory is tight, it uses a shorter context. If memory is really short, the model still loads but runs slowly, and the app tells you plainly: close other apps or choose a smaller model.

It offers small models, and tells you which one to start with#

On an 8 GB computer we suggest Gemma 4 E2B, a 3.3 GB download. It is faster than the larger E4B and less careful. If that is still slow, Qwen2.5 0.5B is a 0.5 GB download that is quick, but it often gets things wrong and isn't reliable with tools, so Pivra Desk never picks it for you. Check its answers. You can keep more than one downloaded and switch in Settings → Model. Choosing a model has the full table.

The app itself is small#

Pivra Desk is about 170 MB to install. The model is the heavy part, and you choose its size.

A cloud model is there when a job needs it#

If a long, multi-step job is too much for a small model on your machine, you can use a cloud model you choose, with your own key. Your chats and memory still stay on your computer, your files stay where they are, and approvals work the same way.

What we haven't measured yet#

Here is the honest part.

The one speed measurement we have published is on a 16 GB machine: on an Apple M2 Pro with 16 GB, Gemma 4 E2B writes about 78 tokens a second (measured 4 October 2026). That is a useful number, but it is not an 8 GB number.

We have not yet published measurements on physical 8 GB hardware. We will measure 8 GB Macs and Windows laptops, and we will publish the results with the machine, the model and the date, the same way we published the 16 GB figure. Until then we will not tell you Pivra Desk is fast on 8 GB, because we haven't shown it.

Our own help pages say 16 GB is comfortable for a model on your computer, and the smallest models run on 8 GB. That is accurate today, and it is the gap we are working to close.

The goal: invisible on your computer#

The bar we set ourselves is higher than "the model loads". On an 8 GB laptop with Chrome and Office already open, Pivra Desk should feel invisible: no long freezes, no fans roaring, no being told to close your other apps.

We are not there yet, and we haven't measured it on 8 GB machines. But it shapes every decision:

• The smallest model that does the job well, not the biggest one that squeezes in.
• Answer without a model when no model is needed. Some questions can be answered with no model at all, which costs almost no memory.
• Predictable over impressive. A steady reply in a few seconds is better than a brilliant one after your computer has stalled.
• No knobs you have to understand. The default should be the right choice for your computer.

Tips for an 8 GB computer today#

If you're on 8 GB right now, these help:

1. Start with Gemma 4 E2B in Settings → Model.
2. Close browser tabs and apps you aren't using before a long task.
3. Use a cloud model for the long jobs if you have a key, and keep the local model for reading and summaries.
4. If it crawls, try Qwen2.5 0.5B for quick, simple questions. It is quick, but it often gets things wrong and isn't reliable with tools, so check its answers. Troubleshooting covers the rest.

Frequently asked questions#

What is Pivra Desk?#

Pivra Desk is a local-first personal AI agent for Mac and Windows. It works across your files, browser and apps using a model on your computer or a cloud model you choose, and asks before consequential actions.

Does Pivra Desk run on 8 GB of RAM?#

Yes: the smallest models are sized for 8 GB, and Gemma 4 E2B is the one we suggest. 16 GB is more comfortable. We haven't yet published measurements on 8 GB machines; when we do, they'll include the machine, the model and the date.

Will it slow down my computer?#

A model uses a lot of memory while it is working. Pivra Desk checks free memory before loading one and picks a context size that fits, and it tells you if memory is too short. Closing unused apps helps.

Which model should I use on an 8 GB laptop?#

Start with Gemma 4 E2B. For quick, simple questions, Qwen2.5 0.5B is smaller and quick, but it often gets things wrong and isn't reliable with tools, so Pivra Desk never picks it for you. Check its answers.

Where does the "most computers in India have 8 GB or less" estimate come from?#

It is our estimate, built from public shipment and price data, because no research firm publishes this split publicly. We hold it with low to moderate confidence on the exact share, and treat it as a design target, not a measured fact.

How much does it cost?#

There is a free plan. Plus is US$6 a month (A$9 in Australia), charged in US dollars outside Australia; taxes may apply. See pricing.

Try it on the computer you have#

Download Pivra Desk for Mac (Apple silicon) or Windows 10 or 11. If it's slow on your machine, tell us what you have. Those reports shape what we measure next.