anyone try setting up a local AI/dev setup with Apple silicon?

[bruce]

Hi.

Looking at dabbling with trying to create test webApps with AI. I've
rad/researched/etc lots of stuff, and the head is still spinning. I've
started to look at actually paying for initial tokens, and then after
seeing large ops recoil at blowing their "token" budgets, decide to
halt this.

So, on a parallel trek, a bunch of information is out about Apple
Silicon and AI dev/coding. Nvidia is still the lead for training, but
for some dev/coding, some say, Apple can be useful.

Is this true. I've looked over some prices, whoa. $12K for a 196Gb
desktop kind of system, and that's going up due to mem!!

The other part of this.. If one were to buy one of these things..
there's a def aftermarket.

So, I'm asking, if anyone here is skilled with your own local setup,
can wee talk, are you willing to share pointers/information -- good or
bad..

My goals are to build some webapps, and some other things, but i can't
go out and hire the devs, and i'm also not looking to do full scale
agentic dev/coding. I'll be the person in the middle so to speak.

thanks

-bruce
bado...@gmail.com

[Jeff Squyres]

Bruce —

You’re asking just in time for our monthly meeting (Wednesday, 6:30pm US Eastern).  Want to come ask this in person?

To unsubscribe from this group and stop receiving emails from it, send an email to kyoss-discus...@kyoss.dev.

-- 

{+} Jeff Squyres

[bruce]

Hi Jeff.

If i can arrange transportation, I definitely will.

I'm next to Shawnee Park, and, well -- you guys aren't!!

But, just in case, where is the location. + building/door + secret handshake!

thanks

[Jeff Squyres]

We have the location and monthly meeting time on https://kyoss.dev/:

• 6:30pm US Eastern time
• 700 N Hurstbourne Pkwy, Louisville, KY, 40222 USA(opens new window)
• First floor: the 700 room
  • You can walk in either entrance to the building; the conference room is on the same floor as the entrance doors.
  • The conference room is easy to find: it has floor-to-ceiling glass walls and glass doors.

Hope to see you there!

[bruce]

thanks much!!

[Andrew Koroluk]

If you want to get started just use whichever machine you have. If your machine has 16gb memory don't download a huge parameter model from HuggingFace; get one of the smaller ones. 

If you're just looking to get started locally there's plenty of free models built for low power (not made for AI or servers) machines. You don't need some monster of a machine just to run some models locally for fun/testing.

------

Andrew Koroluk

[bruce]

hi Andrew.

Thanks for the reply. I'm a bit beyond that stage. I'm at the stage
whereI really need to talk with someone who can say, yeah (or no) with
regard to actually using Apple silicon the Apple m2/m3 ultra type, to
actually do man in the middle code dev processes. I'm not looking to
do autonomous agentic dev.

But i am looking to do something like, pull a good sized Github/repo,
and have the prompts have the AI do an analysis of the app, to then
determine if the functionality I'm looking to take from the given app,
could be modified/rewritten and coupled with other functions to build
what I'm looking for. A kind of "reverse" engineering/Prompt process
to then build/test/review.

I'm currently getting a feel for how this process might work, the
structure of the commands/etc.. But while the free askGem'/Ai is
useful, it has severe limits, and the smaller models are useful for
smaller tests for getting a feel for how this all could work on a
larger system.

I'm not ready to throw cash at the cloud services. But, if I could get
the answers I'm searching for, from an actual person/group, thats
using Apple silicon, than I could have a better idea of which
direction i can go.

I simply don't have the cash to fund a dev team. But I'm intrigued by
the idea that this might work for my needs (or a good portion of my
needs).

thoughts/comments are more than welcome..

thanks
-bruce

[Andrew Koroluk]

 What you're looking to do is called "clean rooming". This is where step one analyzes some software and creates a spec document for it. The second step uses only the spec document (no code/images/artifacts from the original software) to recreate the original or the pieces of the original you want to use.

I've done this before using Gemini Deep Research for part one (I have the $20/mo plan) and then Antigravity with Gemini Pro for part two. You could also use a local deep research tool for part one and an AI IDE with a local model for part two. It can all still just run on your development workstation. I see no need to buy an entire new dedicated machine for this work.

------

Andrew Koroluk

On Mon, Jul 6, 2026, 13:55 bruce <bado...@gmail.com> wrote:

hi Andrew.

Thanks for the reply. I'm a bit beyond that stage. I'm at the stage
whereI really need to talk with someone who can say, yeah (or no) with
regard to actually using Apple silicon the Apple m2/m3 ultra type, to
actually do man in the middle code dev processes. I'm not looking to
do autonomous agentic dev.

But  i am looking to do something like, pull a good sized Github/repo,
and have the prompts have the AI do an analysis of the app, to then
determine if the functionality I'm looking to take from the given app,
could be modified/rewritten and coupled with other functions to build
what I'm looking for. A kind of "reverse" engineering/Prompt process
to then build/test/review.

I'm currently getting a feel for how this process might work, the

structure of the commands/etc.. But j the free askGem'/Ai is

[bruce]

Oh,

Meant to ask/inquire. If you know of someone/anyone playing or working
with this, feel free to fire my contact information to them.

I was testing using win11 vscode+continue.dev extension to
test/connect to my remote ubuntu/ollama/qwen-coder. Something
happened, and the setup borked for the "continue/vscode" process.
Apparently Cursor purchased "continue" so that may have something to
do with it, as the app is being sunset.

I switched to test Cline, but it has issues that I can't quite figure
out to be able to test it.

So, if you know someone, or a group that's into this, I'm more than
willing to shell out some funds to resolve this/these issues. Or,
there might be other open source solutions I haven't run across.

Regards/Thanks!

bruce

On Mon, Jul 6, 2026 at 1:31 PM Andrew Koroluk <koro...@gmail.com> wrote:
>