Joshua Warren
@JoshuaSWarren

Building Remnic (OSS agent memory, ~1mil monthly installs) and mlx-omarchy. MLX on Linux with Apple Silicon. Omarchy M team member.

 Dallas, Texas
joshuawarren.com
Joined May 2011
955 Photos and videos
How it feels doing all the Apple Neural Engine work on my MacBook Pro running Omarchy Linux, writing drivers as I go. It’s actually a lot of fun to drive this way, but I will build a safer, simpler path for those that don’t like driving from the engine compartment!
It's FOSS
@Itsfoss
16h
Very accurate 😂
1
21
1,357
Two more Mac models got their Neural Engine running on Omarchy today. @SiliconForested ran the first M1 Pro test, @ncfrontiersman the first M2 Pro, both came back bit-exact. Also new: a 27B model on my M2 Max now starts answering 25% sooner under Linux. Which Mac is next?
19
7
86
2,899
Omarchy on Apple Silicon makes local AI fast with my MLX and Apple Neural Engine drivers. @dhh inspired me to believe computing should be fun, look good, and run fast. So, I built Coreglass to make inference profiling fun, fast and beautiful. github.com/joshuaswarren/cor…
18
9
255
16,653
more replies
Coreglass turns each run into 1600x900 shareable frames. The synthwave view has a striped 80s sunset and neon grid. Press t to switch to your current Omarchy theme. Standard library Python, no build step. Super+Space, type Coreglass.
1
8
441
I have PRs open to the Omarchy M kernel to add additional instrumentation to make Omarchy the best way to profile your Apple Silicon inference engines. More coming soon!
1
7
310
Got an M1 Pro, M1 Ultra, base M2, M2 Pro or M2 Ultra? Or any M3 or newer Mac? I'm bringing the Apple Neural Engine up on Omarchy and need data from your chip. Run our updated collector under macOS and Omarchy. github.com/joshuaswarren/oma…

omarchy-mlx/docs/contribute-data.md at main · joshuaswarren/omarchy-mlx

MLX-compatible Vulkan and ANE backends for Apple Silicon running Linux. - joshuaswarren/omarchy-mlx

github.com
23
21
198
25,669
My agents are squeezing literally every drop of power from Apple Silicon under Linux this morning: "I'm noticing something odd: VP0b climbs during the pf2048 load phase but then stays surprisingly low, around 0.57–0.69, during the 8.2s prefill while power draw is 10–13W."
2
33
1,840
Big day for local AI on Omarchy with Apple silicon. My driver now boots the M2 Max ANE firmware, loads a compiled model, and passes every check Apple's firmware throws at it. Loading frames now, inference on the M2 Max Apple Neural Engine should follow shortly!
10
10
177
5,258
Trying something new with the agents working on the Apple Neural Engine drivers - may turn it into a skill: "You and your subagents must operate like principal investigators on a major research project pursuing a patent. Maintain detailed engineering notebooks and lab logs."
5
1
16
646
This is after yet another frustrating conversation where something that worked just fine yesterday doesn't work today, and somewhere along the way the boot chain got corrupted and the agent can't tell me where/when/why/how, so it's blindly troubleshooting and wasting time.
1
1
1
200
I don’t have Tobi’s wine cellar, lighting or cable discipline, but I give you my wetbar turned Apple Neural Engine reverse engineering station. If you’re wondering why the MacBook Pros are sitting on each other - agents are using the webcams to monitor output while booting.
tobi lutke
@tobi
Sep 18
Our house in Toronto had this fancy glassed wine cellar, but we are not big wine drinkers. Anyways, it had good climate control so here is my rack cellar.
10
6
103
5,133
Joshua Warren retweeted
Ivan Fioravanti
@ivanfioravanti
Sep 25
Which MLX engines do you actually use? I can't test and eval them all 🤷🏻‍♂️ Community poll: pick up to 5 of 19 (oMLX, LM Studio, Nativ, MLX-Serve, …). Add your chip (M1 to M6) and RAM if you like, to see what each setup prefers. No login, live results, open for 7 days: poll.devocracy.it/mlxengines Please repost to spread the word and don't cheat out there! 🧐

Which MLX engines do you use?

Pick up to five MLX engines you use.

poll.devocracy.it
38
26
117
20,223
Finally cool enough in Dallas to enjoy my coffee and Apple Neural Engine firmware work outside this morning! I need to find a good forest-inspired Omarchy theme - what’s your favorite?
5
1
37
880
No one has consistently brought up the Apple Neural Engine under Linux on anything newer than an M1. Until now. Finally found the breakthrough I needed, and the ANE powers up and responds!
13
15
532
23,505
This is on an M2 Max but I expect the work to apply to all M2 chips. Pretty soon I will have 2 chip families down, 4 more to go!
2
38
1,727
Turns out, from M1 to M2 Apple made a change that requires the hardware video encoder to be powered on before the ANE can be powered on.
2
1
56
2,236
Should be able to release the first alpha version of the ANE driver for M2 under Omarchy tomorrow. In the meantime, some rough notes if you’re interested: github.com/joshuaswarren/oma…

omarchy-ane/docs/t6021-ane-bringup-findings.md at main · joshuaswarren/omarchy-ane

Apple Neural Engine support for Omarchy Linux: DRM accelerator driver and userspace library, M1 T8103 - joshuaswarren/omarchy-ane

github.com
2
1
69
2,203
I won first place in OpenAI's Build for Good dev challenge! Based on Remnic, What Helps Me turns your AI memory into support cards for the people who help you. You approve every word, share only what you choose, and can stop sharing any time. joshuawarren.com/blog/what-h…

What Helps Me won first place in OpenAI's Build for Good challenge — Joshua Warren

What Helps Me, a support passport built on Remnic, took first place in OpenAI's Build for Good challenge. The person approves every card, shares only what they choose, and can stop sharing at any...

joshuawarren.com
3
1
19
589
Having fun watching my agents running on Omarchy on my M1 MacBook Pro drive my M2 via a USB cable to probe every detail of the M2’s Apple Neural Engine. Exciting things coming to mlx-Omarchy today, and not just the massive Qwen3.8 speed ups that shipped in 0.7.2 this morning.
8
6
185
5,346
Load more