What your computer can run
What each amount of memory gets you, the AI models that do the work, and small computers worth buying to leave on. Most people already have enough.
Your computer
Will it run on mine?
The number that decides what runs is memory (RAM). A local AI model has to sit in memory while it works, and WeMemi lets its models use up to half of yours, so everything else on your computer stays quick. Find your number below.
8 GB
Compact modelsYou might have: a MacBook Air with M1, M2 or M3 in its base model, or a budget Windows laptop.
What you get: Life and Mind, capturing and search all work in full. The everyday and reviewer roles run on compact models (Qwen 3 1.7B and Qwen 3 4B) sized to fit. The deep thinker needs 16 GB.
16 GB
Full appYou might have: any MacBook Air with M4, a Mac mini or iMac with M4, a 14-inch MacBook Pro, or most Windows laptops from the last few years, including every Copilot+ PC.
What you get: every feature in WeMemi and all three roles, plus search that understands meaning. The reviewer runs on the compact Qwen 3 4B so everything fits, and one model stays loaded at a time.
See machines ↓18–24 GB
Full-size modelsYou might have: a MacBook Pro with M3 Pro, a 24 GB MacBook Air or Mac mini, or a higher-spec Windows laptop.
What you get: the full-size Gemma reviewer alongside the other two. At 24 GB you can also add the bigger 14–20B models, like gpt-oss 20B, which is our fastest large model.
See machines ↓32–36 GB
No waitingYou might have: a MacBook Pro or Mac Studio with a Max chip, or a Windows gaming PC or workstation.
What you get: everything above, with room to keep two models loaded at once, so moving between quick and deep answers doesn't mean waiting for a model to load.
See machines ↓How much do I have?
Mac: Apple menu → About This Mac. It's the Memory line.
Windows: Settings → System → About. It's Installed RAM.
You don't have to work this out yourself. WeMemi measures your computer on first run and only offers the models that fit.
On Windows, a graphics card helps
Any recent graphics card from NVIDIA, AMD or Intel speeds up answers. Without one, the models run on the processor. Everything still works, it just thinks more slowly.
iPhone and Android
The phone app runs the same memory engine, with its own AI on the phone. It is on its way to the App Store and Google Play. See what it does.
Hardware guide
Buying one to leave on?
WeMemi runs on the computer you already use. But a small computer that stays plugged in and switched on makes a good home base: it keeps your email, calendar and other connected accounts up to date while your laptop sleeps, and your phone always has a computer to sync with. Every pick below is small, quiet when idle, and cheap to leave running.
16 GB
Full appEvery feature and all three roles.
desktop_macMac
Apple Mac mini
M6 · 16 GB memory · 512 GB
Apple's newest Mac mini. Silent, and barely uses any power when idle.
Check price on Amazonopen_in_newSpend less: Mac mini with M4, 16 GB / 256 GB
desktop_windowsWindows
Beelink SER8
Ryzen 7 8845HS · 32 GB memory · 1 TB
Good value, with twice the memory. It has no separate graphics card, so answers come more slowly than on a Mac mini.
Check price on Amazonopen_in_new24–32 GB
Full-size modelsAdds the full-size reviewer, and bigger models like gpt-oss 20B.
desktop_macMac
Apple Mac mini
M4 Pro · 24 GB memory · 512 GB
The Pro chip moves memory more than twice as fast as the standard M4, so the full-size models answer quickly.
Check price on Amazonopen_in_newdesktop_windowsWindows
Minisforum UM890 Pro
Ryzen 9 8945HS · 32 GB memory · 1 TB
A fast processor, plus an OCuLink port if you ever want to add a desktop graphics card for faster answers.
Check price on Amazonopen_in_newPrefer a big-name brand: ASUS NUC 14 Pro+, Core Ultra 9, 32 GB
36 GB+
No waitingKeeps two models loaded at once, so quick and deep answers never wait on each other.
desktop_macMac
Apple Mac Studio
M5 Max · 36 GB memory · 512 GB
Apple's workstation desktop, with the memory speed to run the largest models comfortably.
Check price on Amazonopen_in_newdesktop_windowsWindows
GMKtec EVO-X2
Ryzen AI Max+ 395 · 64 GB memory · 2 TB
Shares its 64 GB between processor and graphics, the way a Mac does. The closest a Windows mini PC gets to a Mac Studio.
Check price on Amazonopen_in_newThese are Amazon affiliate links. As an Amazon Associate, WeMemi earns from qualifying purchases. It costs you nothing extra. Amazon listings often offer several memory sizes, so check the one you pick before you buy. And if the computer you already have is enough, we'd rather you kept it: ask us.
The models
A small team of AIs, all on your computer
WeMemi splits the work between right-sized models instead of one giant one, so a 16 GB laptop can do what would otherwise need a workstation. They download once, straight from the public model registry to your computer. We never see them, or anything you ask them. Each model unloads when it's idle, and you get the memory back. Chat, if you turn it on, talks to model companies on your own key. These local models never do.
Everyday
Gemma 4 E2B · Google
Always on and light: the quick, frequent jobs in the background, and short answers.
~6.6 GB of memory while working. On 8 GB it's Qwen 3 1.7B at ~1.8 GB.
Reviewer
Gemma 4 E4B · Google
Reads what's been captured and proposes what's worth keeping. Nothing is saved until you approve it. It also answers when you ask Mind a question.
~8.9 GB of memory while working. On 8 and 16 GB it's Qwen 3 4B at ~3 GB.
Deep thinker
Gemma 4 12B · Google
Only for the heavy lifting: long summaries and the questions that need your whole history.
~7.5 GB of memory while working
Search
Nomic Embed · Nomic
Search by meaning, not just exact words. Ask about "the trip with Mom" and it finds it.
~0.5 GB of memory while working
Memory figures are measured, not estimated.
Want a different model? Swap one in.Every role has a picker on the Models page. Models too big for your computer stay listed but greyed out, with the reason. Open this for what each one needs.expand_more
| Model | Made by | Memory while working | Runs on |
|---|---|---|---|
| Qwen 3 1.7B | Alibaba | 1.8 GB | 8 GB and up |
| Phi-4 Mini | Microsoft | 2.9 GB | 8 GB and up |
| Qwen 3 4B | Alibaba | 3.0 GB | 8 GB and up |
| Qwen 3 14B | Alibaba | 9.4 GB | 24 GB and up |
| Phi-4 14B | Microsoft | 9.4 GB | 24 GB and up |
| gpt-oss 20B | OpenAI | 11.9 GB | 24 GB and up |
| Qwen 3.8 27B | Alibaba | 18.5 GB | 48 GB and up · Mac |
| Muse Glimmer 30B | Meta | 19.9 GB | 48 GB and up · Mac |
The two largest, Qwen 3.8 27B and Muse Glimmer 30B, need a Mac with 48 GB or more (a MacBook Pro with M4 Pro or M4 Max, or a Mac Studio). They run on Apple's own MLX engine, so they're Mac-only.
Not sure what to buy?
Tell us what you have, or what you're thinking of buying. We'll say plainly what it will run, and if what you have is already enough, we'll tell you so.
Ask usarrow_forward