atrekovas

THIS GUY BUILT A TALKING AI THAT FITS IN YOUR PALM this guy ran an ai companion from a sci-fi novel on a raspberry pi zero over a weekend and that’s already wild but the thread above explains why this is just the start same claude opus 4.5 scored 78% on one harness and 42% on another so that’s a 36 point gap on the exact same model with zero changes to the model itself the model is not the product, the harness is the product and while everyone keeps throwing money at api wrappers this guy already holds a full autonomous ai in his palm with no cloud no subscription no nothing bookmark this and drop a like 👇

YOUR MACBOOK CAN RUN A SELF-IMPROVING AI AGENT RIGHT NOW AND THE SETUP TAKES LESS TIME THAN MAKING COFFEE one terminal command, hermes installs all the dependencies, and by the time your coffee is ready you have a local agent that writes its own skills, remembers everything you teach it and gets smarter every single day while everyone else is still googling “best AI setup 2026” this thing is already running on your desk bookmark this or spend 3 hours looking for it the day you finally get tired of doing everything yourself ↓

PEOPLE STORE NOTES IN OBSIDIAN FOR YEARS AND USE MAYBE 10% OF THEM connected claude to my vault via filesystem mcp and it started pulling connections between notes i wrote months apart in completely different contexts, surfacing ideas i forgot i even captured and building a morning brief around actual projects instead of whatever notifications came in overnight 3 hours a week back from pure routine, blank page problem just gone, more content out the door without adding a single extra working hour obsidian alone is just a graveyard of good ideas obsidian with claude is the first time that graveyard actually talks back full breakdown inside 👇

CLOUD AI ANSWERS IN 3 SECONDS AND FORGETS EVERYTHING LOCAL AI TAKES 15 MINUTES AND REMEMBERS FOREVER hermes running gemma 4 26b through lm studio, compressing context on the fly, processing prompts locally while your data never leaves your machine yes it’s slower, yes it compounds every single day, yes the gap between people who own their AI and people who rent it is about to get very wide full breakdown on how this actually works ↓

HERMES AGENT GEMMA 4 26B ON A MAC MINI NO API KEYS NO MONTHLY BILL NO ASKING PERMISSION TO USE YOUR OWN DATA a $700 box on your desk just replaced a $30/month subscription, runs a 26b model that outperforms what used to need a data center, texts you on telegram and never forgets a single thing you taught it the gap between people who own their ai and people who rent it is about to get very wide full breakdown on how this actually works ↓

@beamnxw genius

A 21-year-old Chinese kid just built a monster farm of Mac Minis > Full server racks > Dozens of M4 Macs stacked like bricks > Running 24/7 at full power While you're bleeding $200–$500/month on Claude/GPT subscriptions… This dude owns his compute completely. No more API bills. No rate limits. No censorship. Just raw local power for pennies in electricity The quiet rebellion is happening right now: Heavy users are canceling expensive subscriptions and switching to local setups ➜ Mac Mini clusters, small server racks, private inference One-time hardware cost > endless monthly rent This is the real "f*ck your subscription" movement in 2026 The ones who own their hardware will move faster, keep their data private, and never get throttled again

THIS IS WHAT 200+ CONNECTED NOTES LOOKS LIKE IN REAL TIME most people store notes and never open them again, this guy connects them and they start thinking on their own every dot is an idea that found a home, every line is a connection that didn’t exist yesterday watch the graph grow and tell me you don’t want this bookmark & like this so your algorithm keeps showing you things worth knowing👇

A 14-year-old from China hacked a FiveM server to understand how it worked - then built his own better one. He opened Claude and described what he wanted - a Lua script with a job system, economy and UI panel. 90 minutes and a finished product on Cfx Marketplace for $389 with zero lines of code written by hand. Then he launched an RP server with a paid whitelist - 200 members at $20 a month is $4,000 recurring every month while he sleeps. GTA 6 drops November 19, 2026. Rockstar is building a creator economy directly into the game - projected annual creator payout is $240 million. GTA 5 made $8.6 billion over 13 years and paid creators $0 because there was no platform. The creators making $1,000,000 a year on GTA 6 in 2028 are already building on FiveM in 2026.

@francescoinweb3 saved this

WANG HAO SAID NO TO $30K GPUS AND YES TO 200 MAC MINIS hundreds of units running 24/7, external fans on every shelf, one dashboard monitoring all of them at once, total monthly cost less than a single claude max subscription nvidia thought they owned the AI infrastructure market, someone built this for $599 a unit and proved otherwise

YOUR OBSIDIAN BRAIN DOESN’T FIRE NEURONS THIS ONE DOES most people set up obsidian and think they built a second brain, this has STDP learning, pattern recognition and agents that feed information into it and pull from it in a cycle, the whole thing just gets smarter on its own obsidian doesn’t do that full guide on how to build this ↓

YOU SPENT 30 MINUTES SETTING UP OBSIDIAN AND IT STILL DOESN'T KNOW YOUR NAME jarvis reads your vault before every conversation and knows your projects and your preferences and your entire history no prompting and no referencing and no digging through folders this is what a second brain looks like when it actually works bookmark & like this if you're tired of talking to an AI that forgets you exist

@beamnxw hahaha yeah...

@atrekovas sad moment fr...

When you hit your token limit on Claude:

THIS ENGINEER HASN’T TOUCHED HIS CODEBASE IN 2 DAYS AND 1000 AGENTS WORKED IT WHILE HE SLEPT he wrote one workflow script and claude code spawned the rest by morning the migration was done, verified, committed and pushed his coworkers are still writing prompts one by one and wondering why nothing ships the only thing separating them is one mental model where orchestrator routes, agents execute and substrate holds the state so no context ever gets poisoned single agent is a bottleneck and multi-agent is how you actually scale full guide below ↓

Jensen Huang (CEO of NVIDIA): “I’ll hire one biological engineer then rent 1,000 AIs” So while your company is posting job listings for 50 engineers, the people who listened to this are already building with one human and a box the size of a Mac Mini Watch him say it, then read the full breakdown below👇

HE RAN A 122B MODEL ON A DGX SPARK AND IT TOOK 69 SECONDS TO SAY HELLO same DGX Spark, same 128GB of unified memory, completely different experience, and the only thing that changed was the model this is what nobody tells you about local AI, the hardware is only half the equation and the model you pick matters just as much as the machine you run it on open source models are getting better every week and the gap between local and frontier is closing faster than most people think bookmark this, the hardware flex is easy, picking the right model is the actual work

THIS ENGINEER BUILT A $4 DEVICE THAT REPLACED THE GUY WHO WALKS ACROSS THE FLOOR TO DELIVER MESSAGES he connected an ESP32 to WiFi, set up MQTT, plugged in an OLED screen went back to his desk by end of day any message sent from anywhere in the building shows up on the display in real time his coworkers are still shouting across the workshop the only thing separating them is one habit: mqtt + esp32 every message = instant delivery. every alert = zero delay. zero missed notices full build breakdown and use-cases ↓

@francescoinweb3 thanks!

With each passing day, NVIDIA is gaining the upper hand in the battle against the Mac Studio, and here’s why it matters to you NVIDIA is redefining the standard for AI resources. Now, in addition to its leadership in the gaming sector, it aims to take the lead in artificial intelligence as well. DGX SPARK packages ✓ 128 GB of memory ✓ An enhanced version of the Blackwell graphics processor ✓ An upgraded data centre network. ✓ The CUBA ecosystem And yet, they’ve managed to fit it all into a box the size of a Mac mini This product is literally designed for leaders in the AI sector, enabling them to utilise such resources within their companies. You are literally buying an AI cluster Developers are now creating prototypes of data centre-scale AI systems. This is changing the landscape of the AI resources market