Kyle Tusing

What I am currently working on… 1. I trade 📈 10+ years, stocks, crypto, options, futures. Disciplined risk/reward. I publish weekly education and setups with a macro read in my newsletter Market Insights & Trading Trends. 2. I build private AI operating systems for businesses. Cloud or locally hosted. Prefer local because it is fully owned, no dependencies or limits. The agents can actually run workflows 24/7, not just answer questions. If you are interested in one of those two things right now 👇🏽 If it's trading: → ktusing34.substack.com If it's AI for your business: → ai.kyletusing.com If it's both, you're in the right place just give me a follow 👊🏽

@jimboot @GIGABYTEUSA Might be able to find slightly cheaper but this was the best option when I last looked amazon.com/dp/B0FXHFBL54?…

@jimboot Haven’t tried mac yet, but this Nvidia GB10 by @GIGABYTEUSA has been great. How large is your team? I have one for my business and a clients business Multiple Hermes agents.

What should I buy if the budget is $5k? Macmini, 3090?? Needs to run Hermes for the team and local model like qwen 27b

@AlexFinn I need more local compute…

I started telling you this in January. I now have 3 Mac Studio 512gbs, 2 Mac Minis, and a DGX Spark. Now you can't even buy Mac Studios over 100gb I told you months ago compute would soon be impossible to obtain as everyone rushes to purchase It's all playing out as I warned

This is your wakeup call. Anthropic just took down Fable 5. It's over. Here's the thing tho: no company or government will EVER be able to take away your local models. There are Opus level models you can run right now on your home GPUs, and nobody can ever stop you from using them This is only the beginning of events like this. Day 1. More government overreach will happen. This will only keep happening more and more as models get closer to AGI Become sovereign. Buy your own compute. Before even that becomes illegal

@AlexFinn I’ve been preaching it!

You can read about my 6+ Hermes agent build, self-healing, plus a clients build that’s in phase 2-3. All running on a Gigabyte DGX spark GB10 box, $0 API cost. linkedin.com/posts/kyle-tus…

Anthropic pulled Claude Fable 5 because it’s a national security threat. Too powerful for non-US citizens to access... Models are getting that capable. This is exactly why I’ve been preaching local, sovereign AI infrastructure. If your business is going to rely on it, own your models. Own your stack. Own your advantage.

@AlexFinn Been saying it… linkedin.com/in/kyle-tusing…

@jasondoesstuff I have a 16GB M2 Pro and I run Codex, Claude, a few Comet Browsers (one with 30+ tabs), Terminals, Email, Slack, Messages, Tailscale, Canva, and a few more all at once and the fans aren’t even on yet

Hey Macbook Pro M4/M5 Max owners, Does running Conductor, Codex, Terminal, an email app, Slack, Chrome, Telegram, Messages, Spotify, and like 19 other random background apps hold up fine? Asking for a friend with a M1 Max laptop. 🙊👀

@hashsriram @NVIDIAAI I’m jealous I can’t lie… maybe it’s time to get another GB10 @nvidia @NVIDIAAI

Dual GB10 cluster @NVIDIAAI

@sudoingX Intermediate moving towards whale status quickly.

anon, how much vram or unified memory is in your rig right now? newbs click honest, whales prove it in the replies 👇

@elvissun Make a feedback API or just a form submission webpage. Initial quality was low on my first run with it but not bad!

so how do you collect feedback from agents? now we have 100s of agents running newsjack.sh, but I have no idea what kind of frictions they are hitting and what's happening in the wild. does having a /feedback api help?

@0xSero Early access? 👀

Guess what? local ai for everyone (: we did it local.ai

@katerleonid No

If someone offered you $1,000,000 today for your entire domain portfolio… Would you say yes? 🤔

It’s all a numbers game in the end

@MichLieben ADS

We built 12 Claude Code skills that run $300K/mo in ads across Google, Meta, and LinkedIn. I'm giving all 12 away, free. 4X ROAS on $1M+ spent. Ivan, our head of growth, built them over 200+ hours running campaigns for our clients. They work the ad platforms from the terminal in plain English: > build custom audiences from your CRM lists > catch creative fatigue before CTR drops > bulk-edit campaigns, bids, and naming in seconds > cut wasted spend on dead Google search terms > audit quality scores and surface keyword gaps Drop them into Claude Code, connect your ad accounts, tell it what you want. It reads the skills and executes. You get all 12 ready to install, plus the 67-min video where Ivan and I walk through how we run them for clients. Reply "ADS" and I'll DM you the repo + the video. Repost to pass it on. Must be following.

@vandotorres Every loop is a prompt but every prompt is not a loop!

A loop is a prompt

@alexocheema @nvidia I’m jealous… one day

.@nvidia gave us all the hardware we need to make local AI awesome. 16 x DGX Spark 3 x RTX Spark 1 x DGX Station 24 x ConnectX-7 Cables 2 x High Speed Switches local dot AI

I’ve been building exactly this stack for SMBs. Agentic OS on owned silicon, Hermes agents orchestrated with watcher/QA agents, shared memory + task logs, and pipelines into a unified vector store so multiple agents can work the same workflow without stepping on each other. I have too much built to cram into one reply, but a lot of the architecture + examples are in my threads here on X and longer breakdowns on my LinkedIn if you want to skim what I’ve shipped. linkedin.com/in/kyle-tusing… I’m currently running my own thing so a strict FT role might be tricky, but I’m definitely open to talking, whether that’s FT, a very involved consulting arrangement, or helping you bootstrap the initial OS and QA layer.

hiring ft ai engineer + build our agentic OS from scratch + codify internal operations with agents + orchestrate QA agents that monitor accuracy and proactively fix issues + deploy hermes agents for team members + architect data pipelines into our global vector store + monitor and automate repetitive workflows + build systems for my personal life + work directly with me on oasis and ventures access to our la office unlimited token budget dm me or reply with what you've built

Running Hermes with a few dozen agents on local + cloud here. What I’d suggest first: 👉 Use a single ‘orchestrator’ agent per session that owns the goal, memory, and task log. 👉 Spin up specialist agents (research, coding, QA, ops) as tools it can call, not separate chats. 👉 Everything they do writes back into the same shared memory + task history so the next session doesn’t start from zero. Multi‑agent can be powerful, but you have to centralize context and decision‑making. If you don’t, you end up with several separate smart agents instead of a team

I’m fully into @NousResearch Hermes. I need your best tips on how to effectively use different agents vs single agent in different sessions. Share with me your best practices!

This is the architectural truth that most businesses won’t hear until it’s too late. Disconnected, ungoverned data + LLM gets you hallucination by design. The governance layer isn’t optional. It’s the whole product. Most SMBs can’t afford Palantir, but the same principles apply at any scale.

“Pointing an LLM at hundreds of disconnected, ungoverned databases gets you a system that hallucinates, is insecure, and unauditable. For something as consequential as our nation’s agricultural data, that is not just useless — it’s dangerous. The Ontology has been the key to delivering AI-enabled technology to every farmer in the country.” At AIPCon 10, the USDA demonstrates how the Ontology now underpins national food supply security.