Vettan

Anthropic just dropped a free course on loop engineering with Fable 5 00:00 - how Claude Code actually works under the hood 05:01 - the agentic loop breakdown 16:21 - the feature 99% of devs never use 19:01 - why voice works better than typing 32:34 - automatic code review with draft PRs 58:39 - using Fable 5 for work beyond code This free course is worth more than most paid Claude Code tutorials Bookmark and watch it later Then read the article below

3 DELL PRECISION TOWER 7810S WITH RTX 3060 12GB CARDS BECAME A PRIVATE AI RACK FOR ALL-DAY LOCAL INFERENCE. not a DGX. not one giant GPU flex. three old Precision towers at the bottom of a rack, each carrying a 12GB RTX 3060, with a Minisforum box acting as the central host. then the boring parts that make it actually usable: Terra Master F8 all-flash NAS for shared storage separate 10G VLAN so model traffic does not fight the normal network custom-length Ethernet cables so the rack does not turn into blue spaghetti this is where local AI starts making sense. not because it beats frontier cloud models. because some work is better when it can run beside your private repos, notes, RAG indexes, Open WebUI sessions, Ollama boxes, llama.cpp experiments, Proxmox nodes, and Tailscale access without another browser tab becoming the whole workflow. three 12GB cards still do not become one clean 36GB GPU. model size, VRAM split, heat, noise, power, and orchestration are still the tax. but as a home-lab AI machine, the payoff is clear: own the boring compute keep private data local rent the cloud only when the model needs to be bigger than the rack

A 19-year-old guy from France turned a dead $599 computer into something that saves him 40+ hours a month and outperforms a $800/month employee His old Mac Mini was just collecting dust. Now it runs a network of autonomous AI agents for pennies Most people paste a YouTube link into Claude, get a summary, close the tab, and forget it a week later. They are paying the "manual trigger tax" every single day He got tired of being the bottleneck. So he stopped Now, he just drops a link into a text file and walks away The Mac Mini pulls the audio, transcribes it for free, and Claude pulls out 5 key ideas, timestamped quotes, and self-test questions. It then automatically links the new note to old ones in his database If it can't find at least 2 real connections, Claude doesn't get to call it done. It goes back and tries again Every article he used to bookmark and never open gets the same treatment every night at 11 PM. It gets summarized, weak arguments are flagged, and fluff is auto-deleted At 6:30 AM, before he's even awake, a super-cheap model (Claude Haiku) picks one old note, writes a real question about it, and sends it straight to his phone Total cost for that entire morning routine? Under $0.01 Total API bill for the month? Less than $15 To hire a virtual assistant to do this manually would cost at least $800 a month He tested every piece by hand first, in a normal chat, running each task 3 times before ever automating it You don't get faster at typing. You just stop being the one pushing the process through every single step The part that used to require a $100/hr developer now just requires knowing what to type

AN NVIDIA ENGINEER'S BUSINESS NETWORK HAS BEEN MAKING 3,000 DECISIONS A DAY AUTONOMOUSLY FOR 90 DAYS. HE HAS NOT LOGGED IN ONCE. THE BRAIN LIVES INSIDE OBSIDIAN 00:03 15 live inputs feed through hidden layers of 27, 22, and 16 neurons. 13 outputs trigger every possible business action. running on TurbinOS 6.6.5 with NVIDIA acceleration the inputs are real-time business signals. inventory, distance, urgency, customer proximity, shift time. the outputs are decisions the network makes without asking anyone. reroute, hold, alert, escalate, hide capacity, flee low-margin regions everything trains inside Obsidian. markdown files as the schema. every business transaction updates a note. every note updates the model. the vault is both the memory and the training set he built it 90 days ago. he has not opened his ERP since. the network handles every decision, every routing choice, every escalation on its own, and posts a daily digest into a single Obsidian note he reads with his morning coffee most operators drown in decisions they should have automated years ago. he watches his business run itself from a vault save this before every operator realizes their business decisions were always a neural network that could have been trained inside Obsidian ↓

@nahid_pro09 scaling to fifteen units just to make fifteen grand sounds like a full time hospitality nightmare

@Vettan0 $997 net per apartment is a solid plan

$15,000/MONTH FROM APARTMENTS HE DOESN'T OWN He bought nothing. Leases long-term, lists nightly, keeps the spread. Do the math: Rent $1,400. Nightly $145 at 74% occupancy. Minus cleaning and utilities. About $997 net from one apartment. Now multiply by the number of apartments.

@misat0x Fr, one cracked lamp and it's straight Ls on the flip game 💀 No cap, maintenance hits different.

@Vettan0 this only looks easy until a guest breaks something

@neheart Yep

@Vettan0 that skylight strip on the roof is a nice touch, the kind of detail that usually gets skipped in auto generated models.

This is the exact moment 3D modeling changed forever Forget manual tracing, tedious extrusions, and spending 8 hours staring at CAD files. Claude Code just casually automated 3D modeling directly from raw blueprint PDFs. Through a revolutionary new SketchUp connector, you just drop in a 3-page construction drawing, and Claude builds a fully realized, render-ready 3D model in under 10 minutes. It doesn't even need the app installed on your PC, it talks directly to the cloud. The exponential leap here is mind-blowing: Old Claude: Gave us clunky, kindergarten-level boxes. Claude Code era: Instantly spits out architectural structures complete with cedar shingles, metal roofing, complex railings, and site topography. Yes, it still misses some micro-details and dimensions, it's not perfect yet. But going from absolute zero to a week's worth of student-level SketchUp skills in a single update? That is unprecedented speed.

@doublenickk np

@Vettan0 thanks for support mate🫂

New alpha?!

@j_mhariee Yes!

A DEVELOPER TURNED HIS SMARTWATCH INTO A REMOTE CONTROL FOR CLAUDE CODE, 1 OBSIDIAN VAULT, 2 TAPS TO APPROVE AN AGENT, AND ZERO NEED TO BABYSIT THE LAPTOP 00:02 a permission request lands on his wrist, two taps later the action is approved, and the agent keeps working without him reopening the terminal. most people still supervise claude code from a desk. he moved the only real bottleneck to his watch, so file edits, commands, and risky actions can be cleared in under 5 seconds. connected to obsidian, the system gets far more useful. claude can read project notes, update markdown, save decisions back into the vault, and only ask for approval when it reaches a protected step. one 300 note vault can support 12 project folders, 40 active tasks, and a full queue of changes while the watch stays the human checkpoint instead of becoming another place to work. that makes the watch a safety gate. routine actions keep moving, dangerous commands stay blocked, and the agent gets freedom without permanent access to everything. bookmark this before approving claude from your wrist becomes normal.

@Yumzlef absolute

@Vettan0 This is AI engineering worth emulating

@ajs6888 Yeah

@Vettan0 手表都开始当代码监工了

@gippp69 Meta bro🔥

THIS DEVELOPER GAVE CLAUDE HIS ENTIRE LOCAL KNOWLEDGE BASE, CUT TOKEN USE BY 80%, AND TURNED A PIXEL DISPLAY INTO A LIVE WINDOW FOR HIS PRIVATE AI 00:01 he points claude at a folder packed with PDFs, notes, and project files, then boots the BUSY Bar while the system starts indexing everything in the background. each document gets split into chunks of 500 words with a 100 word overlap, keeping enough context without sending the same 50 page file into every prompt. a free 90 MB embedding model turns every chunk into 384 numbers. Chroma stores them locally, then searches by meaning instead of waiting for an exact keyword match. one question retrieves only the 3 closest chunks. a 10,000 token manual can shrink to roughly 500 useful tokens, making answers faster while cutting API usage by up to 80%. everything can start with a $5 Claude API balance, while the files, embeddings, and vector database stay on his own machine. each answer returns with the exact source file instead of a confident guess. bookmark this before building your own Claude RAG system.

@0xDezo Alpha

CHINESE DEV MAKES $18,000/MONTH SHIPPING $89 ROCKCHIP BOXES RUNNING HERMES AGENTS THAT REPLACE CHATGPT PLUS BEFORE OPENAI EVEN NOTICES THE CHURN Radxa Rock 5B, RK3588S, 8 cores, 6 TOPS NPU. Hermes 3 8B quantized to fit in 8GB RAM. Wooden end-grain enclosure, IPS touchscreen, USB mic and camera glued on top. OpenAI hemorrhages Plus subscribers as makers realize the $200/month buys you what an $89 board serves offline. Pause at 0:40 — a chat-face UI glowing above an end-grain butcher block housing next to a Dragon Ball Vegeta figure. $89 board. $200/month ChatGPT. Same tokens, different bill. Full build in the video below.

@misat0x Yep

@Vettan0 no headset is the best part

THIS HARDWARE COMPANY CREATED A HOLOGRAPHIC 3D SCREEN THAT LITERALLY LETS YOU SEE INSIDE A PATIENT'S BRAIN If you are still trying to analyze complex medical scans or diagnostic imaging on flat, two-dimensional screens, you are looking at data through an outdated lens. This company, Illumetry, just completely changed the game by building a literal holographic 3D monitor setup that lets you interact with a human brain scan right on a desktop. Tbh, watching how it works in practice is pretty mind-blowing. The setup shows a specialized display tracking a physical frame that someone holds in their hands. As they move the frame closer, tilt it, or rotate it, the monitor instantly shifts the angle of the 3D head scan, showing a highly detailed cross-section of the brain tissue beneath it. There are no clunky virtual reality headsets or heavy goggles required. The screen itself handles the deep spatial computing, calculating the exact perspective in real-time to create genuine depth perception out of thin air. It is an incredibly strong proof of concept for localized medical engineering, giving specialists a way to dig into raw diagnostic data naturally And what if we added AI to all of this as well?