Looks like they just added homebrew tap to the instructions

There's a very similar afm CLI that can be installed via Homebrew.

https://github.com/scouzi1966/maclocal-api

good idea

yeah, it is super limited but also you can now do

  cmd(){ local x c r a; while [[ $1 == -* ]]; do case $1 in -x)x=1;shift;; -c)c=1;shift;; *)break;; esac; done; r=$(apfel -q -s 'Output only a shell command.' "$*" | sed '/^```/d;/^#/d;s/^[[:space:]]*//;/^$/d' | head -1); [[ $r ]] || { echo "no command generated"; return 1; }; printf '\e[32m$\e[0m %s\n' "$r"; [[ $c ]] && printf %s "$r" | pbcopy && echo "(copied)"; [[ $x ]] && { printf 'Run? [y/N] '; read -r a; [[ $a == y ]] && eval "$r"; }; return 0; } 

cmd -c show disk usage sorted by size

cmd -x what process is using port 3000

cmd list all git branches merged into main

cmd count lines of code by language

without calling home or downloading extra local models

and well, maybe one day they get their local models .... more powerful, "less afraid" and way more context window.

In Apple’s defense, they did make it do something borderline useful while targeting a baseline of M1 Macs with 8 GB of RAM (and even less in phones).

just german for apple, cause reasons

> False, it creates consumer demand for inference chips, which will be badly utilised.

I think the opposite is true. Local inference doesn't have to go over the wire and through a bunch of firewalls and what have you. The performance from just regular consumer hardware with local, smaller models is already decent. You're utilizing the hardware you already have.

> The performance limitations are inherent to the limited compute and memory.

When you plug in a local LLM and inference engine into an agent that is built around the assumption of using a cloud/frontier model then that's true.

But agents can be built around local assumptions and more specific workflows and problems. That also includes the model orchestration and model choice per task (or even tool).

The Jevons Paradox comes into play with using cloud models. But when you have less resources you are forced to move into more deterministic workflows. That includes tighter control over what the agent can do at any point in time, but also per project/session workflows where you generate intermediate programs/scripts instead of letting the agent just do what ever it wants.

I give you an example:

When you ask a cloud based agent to do something and it wants more information, it will often do a series of tool calls to gather what it thinks it needs before proceeding. Very often you can front load that part, by first writing a testable program that gathers most of the necessary information up front and only then moving into an agentic workflow.

This approach can produce a bunch of .json, .md files or it can move things into a structured database or you can use embeddings or what have you.

This can save you a lot of inference, make things more reusable and you don't need a model that is as capable if its context is already available and tailored to a specific task.

> What makes you think that?

Looking at actual users of LLMs

> What makes you think that?

The fact that today's and yesterday's models are quite capable of handling mundane tasks, and even companies behind frontier models are investing heavily in strategies to manage context instead of blindly plowing through problems with brute-force generalist models.

But let's flip this around: what on earth even suggests to you that most users need frontier models?

Qwen3.5 has tool calling, so you can give it a wikipedia tool which it uses to know what happened in Tiananmen Square without issues =)

Cool, I always wanted to invade Belgium. Maybe if my plan is good, I could run a successful gofundme?

Oh does llama.cpp use MLX or whatever? I had this question, wonder if you know? A search suggests it doesn’t but I don’t really understand.

Did you get any insights about yourself from this process? I am thinking of doing the same

I agree with you in the sense that if you tried to take any model right now and cram it into an iphone, it wouldnt be a claude-level agent.

I run 32b agents locally on a big video card, and smaller ones in CPU, but the lack there isn't the logic or reasoning, it is the chain of tooling that Claude Code and other stacks have built in.

Doing a lot of testing recently with my own harness, you would not believe the quality improvement you can get from a smaller LLM with really good opening context.

Even Microsoft is working on 1-bit LLMs...it sucks right now, but what about in 5 years?

But the OP is correct -- everything will have an LLM on it eventually, much sooner than people who do not understand what is going on right now would ever believe is possible.

It depends. If they're using a small/medium local model as a 1:1 ChatGPT replacement as-is, they'll have a bad time. Even ChatGPT refers to external services to get more data.

But a local model + good harness with a robust toolset will work for people more often than not.

The model itself doesn't need to know who was the president of Zambia in 1968, because it has a tool it can use to check it from Wikipedia.

Any citations? Because that was my impression, too. I want frontier model performance for my coding assistant, but "most users" could do with smaller/faster models.

ChatGPT free falls back to GPT-5.2 Mini after a few interactions.

> unfortunately, this is not the case

Most users are fixing grammar/spelling, summarising/converting/rewriting text, creating funny icons, and looking up simple facts, this is all far from frontier model performance.

I've a feeling that if/when Apple release their onboard LLM/Siri improvements that can call out if needed, the vast majority of people will be happy with what they get for free that's running on their phone.

"Hey dingus, set timer for 30 minutes"

eh, its weird how thetech world wants to build trillions of data centers for...what, escapingthe permanent underclass?

I think what "need" youspeak of is a bit of a colored statement.

Brilliant. Hope to see you in the App Store!

I'd like to coin the term "user agent" for this

Indeed. Data centers have so many ways and reasons to be much more energy-efficient than local compute it's not even funny.

As things stand today even when doing research tasks, time spent by model is >> than fetching websites. I don't see it changing any time soon, except when some deals happen behind the scenes where agents get to access CF guarded resources that normally get blocked from automated access.

While data centres indeed have awesome internet connectivity, don’t forget the bandwidth is shared by all clients using a particular server.

If you have 100 mbit/sec internet connection at home, a computer in a data centre has 10 gbit/sec, but the server is serving 200 concurrent clients — your bandwidth is twice as fast.

> For many businesses it will still make sense to have more powerful models and to run them centralized in a datacenter.

Agree, and I think of it this way: for a lot of businesses, it already makes sense to have a bunch of more powerful computers and run them centralized in a datacenter. Nevertheless, most people at most companies do most of their work on their Macbook Air or Dell whatever. I think LLMs will follow a similar pattern: local for 90% of use cases, powerful models (either on-site in a datacenter or via a service) for everything else.

I agree. I can totally see in the future that open source LLMs will turn into paying a lumpsum for the model. Many will shut down. Some will turn into closed source labs.

When VCs inevitably ask their AI labs to start making money or shut down, those free open source LLMS will cease to be free.

Chinese AI labs have to release free open source models because they distill from OpenAI and Anthropic. They will always be behind. Therefore, they can't charge the same prices as OpenAI and Anthropic. Free open source is how they can get attention and how they can stay fairly close to OpenAI and Anthropic. They have to distill because they're banned from Nvidia chips and TSMC.

Before people tell me Chinese AI labs do use Nvidia chips, there is a huge difference between using older gimped Nvidia H100 (called H20) chips or sneaking around Southeast Asia for Blackwell chips and officially being allowed to buy millions of Nvidia's latest chips to build massive gigawatt data centers.

This seems to be somewhat similar to web browsers.

I could see the model becoming part of the OS.

Of course Google and Microsoft will still want you to use their models so that they can continue to spy on you.

Apple, AMD and Nvidia would sell hardware to run their own largest models.

You can have viable business model around open weight models where you offer fine tuning at a fee.

You can always delegate sub agents to cloud based infrastructure for things that need more intelligence. But the future indeed is to keep the core interaction loop on the local device always ready for your input.

A lot of stuff that we ask of these models isn't all that hard. Summarize this, parse that, call this tool, look that up, etc. 99.999% really isn't about implementing complex algorithms, solving important math problems, working your way through a benchmark of leet programming exercises, etc. You also really don't need these models to know everything. It's nice if it can hallucinate a decent answer to most questions. But the smarter way is to look up the right answer and then summarize it. Good enough goes a long way. Speed and latency are becoming a key selling point. You need enough capability locally to know when to escalate to something slower and more costly.

This will drive an overdue increase in memory size of phones and laptops. Laptops especially have been stuck at the same common base level of 8-16GB for about 15 years now. Apple still sells laptops with just 8GB (their new Neo). I had a 16 GB mac book pro in 2012. At the time that wasn't even that special. My current one has 48GB; enough for some of the nicer models. You can get as much as 256GB today.

I'm expecting someone to come up with an LLM version of the Coral USB Accelerator: https://www.coral.ai/products/accelerator

Just plug in a stick in your USB-C port or add an M.2 or PCIe board and you'll get dramatically faster AI inference.

That's completely not true. LLM on device would use MORE electricity.

Service providers that do batch>1 inference are a lot more efficient per watt.

Local inference can only do batch=1 inference, which is very inefficient.

Apple via customers paying for the whole solution ( eg a laptop that can run decent local models )?

I think Apple had something in the region of 143 billion in revenue in the last quarter.

Not saying it will happen - just that there are a variety of business models out there and in the end it all depends on where consumers put their money.

> LLMs are far more efficient on hardware that simultaneously serves many requests at once.

The LLM inference itself may be more efficient (though this may be impacted by different throughput vs. latency tradeoffs; local inference makes it easier to run with higher latency) but making the hardware is not. The cost for datacenter-class hardware is orders of magnitude higher, and repurposing existing hardware is a real gain in efficiency.

I'm actually not sure that's true. Apart from people buying the device with or without the neural accelerator, the perf/watt could be on par or better with the big iron. The efficiency sweet-spot is usually below the peak performance point, see big.little architectures etc.

> Sorry to shatter your bubble, but this is patently false, LLMs are far more efficient on hardware that simultaneously serves many requests at once.

You might want to read this: https://arxiv.org/abs/2502.05317v2

Well this is an article about running on hardware I already have in my house. In the winter that’s just a little extra electricity that converts into “free” resistive heating.

When local LLMs get good enough for you to use delightfully, cloud LLMs will have gotten so much smarter that you'll still use it for stuff that needs more intelligence.

Yep. People were claiming DeepSeek was "almost as good as SOTA" when it came out. Local will always be one step away like fusion.

It's just wishful thinking (and hatred towards American megacorps). Old as the hills. Understandable, but not based on reality.

We are 100% there already. In browser.

the webgpu model in my browser on my m4 pro macbook was as good as chatgpt 3.5 and doing 80+ tokens/s

Local is here.

Local RTX 5090 is actually faster than A100/H100.

You're assuming throughput sets the value, but offline use and privacy change the tradeoff fast.

Looking at downvotes I feel good about SDE future in 3-5 years. We will have a swamp of "vibe-experts" who won't be able to pay 100K a month to CC. Meanwhile, people who still remember how to code in Vim will (slowly) get back to pre-COVID TC levels.

OK thanks! That's helpful. I ignorantly assumed simpler prompt == faster first response.

Indeed. Qwen doesn’t just second guess itself, it third and fourth guesses itself.

Thanks! I assumed simpler == faster, but my ignorance is showing itself.

I am using the model they recommended in the blog post - which I assumed was using MLX?

You can also use OpenWebUI locally which should give you a nice friendly UX once you set it up.

It's odd no manufacturer jumped on this wagon to offer a competitive alternative.

Note though that that a MAX 395 has half the memory bandwidth of a M4 Max chip, and the memory bandwidth is going to be the biggest limiting factor, so you'll likely be getting around half the tokens/second with that Framework Desktop.

You can, but the quality sucks.

Local LLMs don't make sense for most people compared to "cloud" services, even more so for coding.

You can run Qwen3.5-35B-A3B on 32GB of RAM sure, although to get 'Claude Code' performance, which I assume he means Sonnet or Opus level models in 2026, this will likely be a few years away before its runnable locally (with reasonable hardware).

I'm reading "more than 32GB of unified memory" to mean at least a 36 GB model.

On my M4 Pro MLX has almost 2x tok/s

i've found llama.cpp (as i understand it, ollama now uses their own version of this) to work much better in practice, faster and much more flexible.

Can you please write about your hardware?