George M

@jpudysz did you try herdr (and herdr-web with the mobile app just released)?

Ok Orca with Tailscale looks good. Will test if for few days. I can manage my agents from phone and MacBook 🤩

What’s the best agent orchestration tool for running Codex or Claude sessions on my second Mac over SSH?

@mgoin_ @TheZachMueller Michael, how mi355x compares to gb300, do you have any experience with amd gpu?

@TheZachMueller Let me know if you want any tips! I'm scaling up on multi-node GB300 DSpark training now

GLM 5.2 DSpark update! The full Speculators training run is well underway and we have the epoch-1 checkpoint ready for your GPUs using vLLM nightly: huggingface.co/RedHatAI/GLM-5… This improves upon the speedup from the preview checkpoint by another 1.5-2x. Stay tuned for more!

@AlemTuzlak maybe x.com/NathanFlurry can help us :)

@luckymoooon Would you like to PR it?

Yesterday I triaged a bug on TanStack AI using a new upcoming feature in TanStack AI. Gotta triage it with 16 different scenarios though to thoroughly test it out 😅 We live in some crazy times.

@amanvarshney01 Hey Aman, any plans with wasm based sandboxes like agent-os?

Better T Stack now supports Docker deployment 🐳 Pick docker for web and/or server → Dockerfiles + one docker-compose.yml for your frontend, backend & database Try Now! bun create better-t-stack@latest

@CanadaHonk @svnholup so, warming 1M js threads took 7s only?

@svnholup yep! runtime is go-like (api is to be decided)

1 million threads alive simultaneously using 15.9gb: 16kb/thread, 0.7 μs/start avg. same vm/gc/etc. zero copy. (extremely early!)

@gaeldelalleau @__smiz @Kog__AI @grok any plan to add at least 8B llms, qwen 27B support would be even better

@__smiz @Kog__AI @grok no, our code is written in C++ and PTX/CDNA assembly and optimized toward very specific GPU types

🚀 Launch today: Kog generates 3,000+ output tokens/s per single request, on standard datacenter GPUs. We are bringing real-time LLM inference to hardware that companies already run in production. The speed previously associated with purpose-built silicon is now delivered on NVIDIA H200 and AMD MI300X. Today, we are opening our Tech Preview with a 2B coding model, with large frontier MoE support coming next. Try our Playground → playground.kog.ai 💥 Why that matters, and how we did it → blog.kog.ai/real-time-llm-… 📖 Monokernel deep dive → blog.kog.ai/building-a-sin… 📖 Delayed Tensor Parallelism research → blog.kog.ai/delayed-tensor… read the thread 👇

@buildwithdjdev @StrengthClub1 Thank you, what's your opinion on delivering xcode sdk as portable wasm modules(or you know a better solution to reach portability) to avoid containers? This company is using cheerx tech to run virt x86 apps in browser labs.leaningtech.com/blog There is also github.com/rivet-dev/agen…

@luckymoooon @StrengthClub1 I have a Windows box arriving today. Will be configuring Linux on it too. Better Windows/Linux host support coming to SimDeck soon!

SimDeck is now available on App Store! Access iOS simulators and Android emulators & test the apps your agents are making remotely from your machine. Works over LAN & Tailscale support built-in. $ npm i -g simdeck@latest && simdeck pair Install the app. Scan QR code. Go!

@buildwithdjdev @StrengthClub1 It's working only on macs? Can you provide info how to config it in scenarios like this dev.to/ianito/how-to-…

@StrengthClub1 Works for anything you run on the Simulator :)

@josevalim José, can you try riveter project from github, a combination from ralph loops and jj versioning created by rivet author? may be response to what you're looking for

Additional context: I have seen many people complaining about the complexity of worktrees, so I am looking for alternatives for tidewave.ai. Running multiple containers is a good call out which I missed. Maybe we should explore a more general solution to the problem.

@mitchellh @misterfitzie Hey Mitchell, did you see the mojo x8 performance against rust in a 50k loop object creation/destruction, written in february 24 on the mojo creator's website modular.com and that mojo reached v 1.0 beta last week, who knows, maybe rust is not the bun last destination

@misterfitzie Nope, not at all what I said. Totally respect what Bun did.

It isn't unexpected that the focus of the Bun Rust rewrite is on the anti-Zig side more than anything, since the internet loves to hate. What is unexpected and unfortunate is that leadership within Bun hasn't tried to steer the conversation away from that at all. There are so many positive and interesting takeaways from this and I'm not really seeing any of them pushed as the primary message. A positive thing that hasn't been talked about at all is how far Bun came thanks to Zig. And even if you dump it now, its meaningful for how good Zig was to even build a product to this point and impact by any metric. I would've loved to see anyone in leadership say this. On the interesting side is how fungible programming languages are nowadays. Programming languages used to be LOCK IN, and they're increasingly not so. You think the Bun rewrite in Rust is good for Rust? Bun has shown they can be in probably any language they want in roughly a week or two. Rust is expendable. Its useful until its not then it can be thrown out. That's interesting! There's been a lot of talk about memory safety and no doubt Rust provides more guarantees than Zig. But I'd love to see a better analysis of why Bun in particular suffered so much rather than take the language-blame path. How could engineering as a practice been more rigorous to prevent this? What were the largest sources of crashes other programs should watch out for? How does Rust prevent them? How could Zig theoretically prevent them? That's interesting. I know the official blog post hasn't come out yet from Bun. But they're smart enough to know that that PR would stir up controversy the moment it opened, or they should've been. And plenty in the company have been tweeting and writing about it. Its somewhat telling to me in various dimensions what they chose to talk about first. I tend to think I'm pretty good at corporate PR/comms (especially when it comes to developer audiences) and I think appealing to the negative is never the right long term strategy; it does work to get short term eyes though.

@buildwithdjdev wasn't the m4 max with 48 GB but 40 core GPU and 546 GB/s bandwidth a better option at 3000$ as seen on the apple refurbished store?

I’M READY!! It’s coming early omg

@joelteply @spiritbuun @emreceng_ Hey Joel, any opinion on the modular/max engine based on the mojo lang, in many scenarios is superior to cuda and other engines/sdks, also have support for apple silicon since september and rdna consumer gpus since a week or so

@spiritbuun @emreceng_ Fast. My stuff must run on MacBook m1, multimodal, for all models, all persona, so I’m obsessed with performance. Porting qwen 3.5 27b to MacBooks now. I’m a C++ dev converted to rust.

TurboQuant CUDA for llama.cpp: 3.5x KV cache compression that BEATS q8_0 quality (-1.17% PPL) 99.6% prefill speed, 97.5% decode 128K context on RTX 3090 24GB, Q6 Qwen3.5 27B github.com/spiritbuun/lla…

@fantparl @0xSero how they are supposed to have better bandwidth - will come with upcoming GDDR7 40 gbps or more than 512 pins?

@0xSero It's doing better than expected. This screams BUY A 5090! But I'm waiting to see the rtx 50 supers with better bandwidth memory. Though rumours, no 32gb option.

A 27B model is #2 on pinch-bench You’d need 150,000$ in GPU hours to train this from scratch (base + post training) Basically 1-2 weeks over 256 H100s That is not unreasonable, you’d need 540B tokens for pre-training and a bit more for post training. None of this is crazy

@haknkskn @xgodil @Kimi_Moonshot what's best - cline, roo code or kilo code?

@xgodil @Kimi_Moonshot

You share, we care. Kimi Code is now powered by our best open coding model, Kimi K2.5 🔹 Permanent Update: Token-Based Billing We’re saying goodbye to request limits. Starting today, we are permanently switching to a Token-Based Billing system. All usage quotas have been reset to give you a fresh start. 🎉 Limited-Time Event (Ends Feb 28th) To celebrate this upgrade, we are unleashing full power! - 3X Quota: Enjoy up to 3 times the usage limits. - Full Speed: No throttling. No purchase limits. Just code. Jump in now and build something amazing!

@xenonwellz @AmanVirk1 how about elysia, which one would you choose

@AmanVirk1 Not yet, I’ll try and I’ll definitely raise an issue if I find one

The love keeps growing and I love that 😍

@itsjustmarky @0xSero how fast is glm-4.7-flash running on strix-halo, did you run it?

@0xSero Rocm has been finally making some progress only recently though. I’ve had my Strix Halo for a short time but it has been very noticeable.

Just got 500$ in credits on HotAisle I am excited to try this out. ROCm seems like a good alternative to Nvidia. Will benchmark and explore hotaisle.xyz

@heiwen @saltyAom did you try tanstack start, how does it compare to rr7?

Bun’s full-stack dev server makes for a nice demo, but the lack of SSR + hydration makes it hard to recommend for most production workloads. Personally, I’ve grown to really like React Router 7 (formerly Remix 2) in framework mode with its data loader pattern. It covers the majority of data-fetching needs out of the box and removes the need for reactive query libraries like TanStack Query in 95% of my app.

Claude made simple CRUD with Elysia, Better Auth, Drizzle and Bun Fullstack Dev Server with React, Tailwind and React Query 90% of the code is by Claude and GLM 4.7 It still doesn't know how to do something, bad at React Query and too much abstraction Still need hand holding

@TobiM can you check the glm-4.7 with the same promt to compare, the Pro plan, can be purchased at $144/year with 200m tokens every 5 hours

Claude Code just one-shotted a migration of a fairly complex Electron app to Tauri in about 20min... Planning mode with a few lines prompt -> Success. Incredible.

@ivanfioravanti @s_j_thapa the same thing for Ultra, apple need to double the bandwidth to compete with RTX PRO 6000 with 1.8 TB/s at 7000$ with 96 GB VRAM, not to mention risc-v cips like x.com/jimkxa/status/… or tenstorrent.com/hardware/black… (with gpddr7 will have 1.3TB/s bandwidth, x2.5 more) at 1400$

@s_j_thapa I hope 550 GB/s Max and 1.100 GB/s Ultra 🤞🏻

If this is true, M5 Ultra will be M3 Ultra x2! 🔥 wccftech.com/m5-max-gpu-sco…