English
Thanh @ B.ARMY Ventures
97 posts
Thanh @ B.ARMY Ventures
@techcomthanh
Founder of https://t.co/aE3kNrNB8C, Blockchain Fund focusing on #Ai and #Depin. A Value Investor & 🏌 lover
Tham gia Şubat 2018
201 Đang theo dõi1.8K Người theo dõi
you should try gstack if you haven't yet
Dan Nelson@danimalnelson
@garrytan Gave this a spin last night. Super powerful.
English
English
If you have your OpenClaw working 24/7 using frontier models like Opus, you're easily burning $300 a day.
That's $100,000 a year.
I have 3 Mac Studios and a DGX Spark running 4 high end local models (Nemotron 3, Qwen 3.5, Kimi K2.5, MiniMax2.5). They're chugging 24/7/365. I spent a third of that yearly cost to buy these computers
I'll be able to use them for years for free
On top of that they're completely private, secure, and personalized.
Not a single prompt goes to a cloud server that can be read by an employee or used to train another model
I hope this makes it painfully obvious why local is the future for AI agents. And why America needs to enter the local AI race.
English
English
🚀 Introducing the Qwen 3.5 Small Model Series
Qwen3.5-0.8B · Qwen3.5-2B · Qwen3.5-4B · Qwen3.5-9B
✨ More intelligence, less compute.
These small models are built on the same Qwen3.5 foundation — native multimodal, improved architecture, scaled RL:
• 0.8B / 2B → tiny, fast, great for edge device
• 4B → a surprisingly strong multimodal base for lightweight agents
• 9B → compact, but already closing the gap with much larger models
And yes — we’re also releasing the Base models as well.
We hope this better supports research, experimentation, and real-world industrial innovation.
Hugging Face: huggingface.co/collections/Qw…
ModelScope: modelscope.cn/collections/Qw…
English
English
@techcomthanh @VitalikButerin @Alibaba_Qwen i think it does, also probably irrelevant for his use case
English
English
Thanks @tim_cook for absorbing memory costs. Stacking macs is still the cheapest way to run frontier AI.
Alex Cheema@alexocheema
Nobody is talking about @apple keeping prices the same for the 128GB MacBook Pro. There has been no price increase in response to surging memory prices. Everyone is talking about the boost in compute, speeding up prefill by 4x. This is cool but practically it’s not that big of a deal. Why? Because on your own computer, most apps/tools using LLMs are going to get high kv cache hit rates - that means as a user you only experience slow prefill once. kv cache can be persisted to disk and loaded at 6GB/s. Most time in LLM inference is spent on decode, which is memory bandwidth bound. It’s still great for image/video generation, high batch LLM inference and fine-tuning, which are compute bound. We should see huge speedups there. Apple’s AI strategy is on-device LLMs and here, memory is the name of the game, not FLOPS. Expect the same for M5 Pro/Max Mac Mini and M5 Ultra Mac Studio. That means 512GB M5 Ultra at 10k! @tim_cook is a supply chain genius.
English
English
@alexocheema @exolabs Can we run multiple inferencing requests to a same model using exo?
English
Just wanted to say we look at every bug report sent through the @exolabs app.
I see a lot of reports coming through today and really appreciate it as it helps us a lot to make exo better.
We have some really cool features shipping this month.
English
🚨 Big update to Marketing Skills for Claude Code
26 skills. 29 tool integrations. Better performance across the board.
Here's everything that changed ↓
English
@_StanGirard You dont actually need hidden reversed engineering for Claude Code CLI. Claude agent SDK can do it smoothly. We also build the complete solution for working on different devices at claude.ws
English
I was burning $200/day on agent API calls.
Then I realized: I already pay $200/month for Claude Code Max.
So I reverse-engineered its protocol.
Now I spawn agents via REST API, monitor them from a dashboard, and pay nothing extra.
OSS 👇
github.com/The-Vibe-Compa…
English
@_StanGirard We absolutely can use subscription with it. That is what we have already doing with claude.ws
English
To everyone saying you can get already do this with the sdk. Yes you can almost, however you can’t use your subscription with it.
English
@ollama Ollama + Claude code CLI + Claude WS + Cloudflare Tunnel would be a great combo to work anywhere, syncing on any device
English
ollama launch is a new command in Ollama 0.15 to run Claude Code, Codex, Droid and OpenCode with Ollama!
GLM 4.7 Flash is now optimized to use much less memory for longer context lengths (64k+).
Need additional hardware? Ollama's cloud offers GLM 4.7 with full precision and context length.
English
@NirDiamantAI Pls also add github.com/Claude-Workspa… for webbased claude code deeply integrated, and with kanban
English
Claude Code power users, you’ll want to see this.
There’s a public repo that’s basically a complete operating system for Claude Code: agents, skills, hooks, commands, rules, MCP configs, all wired together and ready to plug in.
Instead of guessing how to structure your setup, you can study or adapt a full opinionated configuration that’s already been battle tested in real projects.
Repo: github.com/affaan-m/every…
English
My personal experience is that the way we build products is changing at lightning speed. As someone who codes part-time (since I also handle the business side—though I’m still passionate about coding) being able to ship like this (check out that empty To-Do list!) feels AMAZING even looking back at it myself.
Most of these tasks were actually done on my phone.
I’m also throwing in a screenshot of how I plan marketing using the Mkt Kit from Claudekit.
To save everyone from having to DM me, I’ll drop the link to the app I’m working on in the comments
English
Caude Code bros can use this tool to 'vibe code' anywhere. Fork it from github.com/Claude-Workspa…, deep integration for Claude Code, then use a Cloudflare Tunnel to access the ClaudeWS web interface.
English
@milesdeutscher Claude Code bros can use this tool to 'vibe code' anywhere. Fork github.com/Claude-Workspa…, it's a deep integration for Claude Code, then use a Cloudflare Tunnel to access the ClaudeWS web interface.
Happy Vibe Coding, anytime
English
If you're building with Claude Code, you'll want to bookmark this site.
A full agent marketplace of 60,000+ Claude Skills that are ready for use now.
https:// skillsmp. com/
English
@claude_code Claude Code bros can use this method to 'vibe code' anywhere. Fork github.com/Claude-Workspa…—it's a deep integration for Claude Code—then use a Cloudflare Tunnel to access the ClaudeWS web interface.
Happy Vibe Coding, anytime
English
Take advantage of @claude_code plugins and skills.
Frontend-design
Doc-coauthoring
Skill-creator
Code-simplifier
Boris Cherny@bcherny
We just open sourced the code-simplifier agent we use on the Claude Code team. Try it: claude plugin install code-simplifier Or from within a session: /plugin marketplace update claude-plugins-official /plugin install code-simplifier Ask Claude to use the code simplifier agent at the end of a long coding session, or to clean up complex PRs. Let us know what you think!
English