Working notes and sources for every video. Each page lists every figure used, where it came from, and what could not be verified.

The channel: @LocallyExplains

Every video

• The Best Local AI Models For A 48GB VRAM Setup
• The Only GPUs Worth Buying For Local AI (October 2026)
• The VRAM You Need For Every Local AI Model (October 2026)
• Don’t Upgrade Your 8GB GPU Until You Try These Models
• RX 9060 XT Runs 27B AI For $270 Less Than Nvidia
• The RTX 4090 Wins Until Your Model Hits Its 24GB Wall
• Bonsai 2 Packs 27B AI Into a File Smaller Than 6GB
• Four V100s Nearly Matched The RTX 5090. Here Is Why
• Mac Mini M6: The 32GB Upgrade Changes Local AI
• The $1,699 Strix Halo Deal Is Gone For Now (128GB)
• The $13,999 DGX Spark Pair Reads 3x Faster Than A Mac
• The 8GB GPU Trap For Real Time Voice AI (RTX 5060)
• Four R9700s Cost $6,800 Before You Even Get A PC
• Run Qwen 27B For $350, Or Fit 284B In A Mini PC
• Six MI50s Hit 192GB Of VRAM For $3,774 With One Catch
• Your 16GB GPU Has a 27B Model Hiding Inside Today
• Don’t Pay $1,690 For The RTX 5080 Until You See This
• Don’t Preorder NVIDIA’s First 128GB CUDA Laptop Yet
• An 8GB Mac Is The Wrong Buy For Local AI Today
• Claude Grade Coding On A Cheap 12GB GPU (Local AI)
• Fine Tune 8B AI On A $480 RTX 3060 (One Catch)
• October 2026 Local AI Buyer’s Sheet (Members)
• 16GB Is Enough For Local AI Until You Hit These Limits
• Kimi K3 At Home Costs 240 Months Of Claude Max!
• The RTX 3090 Trap That Costs Local AI Buyers $1,600
• RTX 4090 48GB Mods Have One Expensive Catch For Buyers
• Your Local AI Lab Costs More Than The GPU (Three Builds)
• Your $200 AI Plan Can Get Worse. Your Local AI Can’t
• Don’t Buy a Loud GPU Tower for Local AI All Day
• Intel’s $799 24GB GPU Takes On The Used RTX 3090
• Your 12GB GPU Can Run 125B AI, If You Own This
• 128GB Local AI Has One Model That Makes It Worth It
• Four RTX 3090s Run 100B AI Fast, But There Is a Catch
• The Best Local AI Machine For $2,000 Has A Catch
• RTX 5090 Can Replace Claude. What Is The Catch?
• The Real Reason Developers Keep Running AI Locally
• 7 Qwen 3.8 models, and the biggest is not the best
• Local AI Hit 150 Tok/s. So Why Does It Still Feel Slow?
• Qwen 125B On Strix Halo Runs, But At What Speed?
• The Cheapest Way to Run 100B AI Has a Catch Today
• The Mac Mini Has a Local AI Limit Nobody Sees Coming
• $500 for local AI and one GPU changes everything
• 32GB VRAM Can Run These Models, But One Stands Out
• Two Cheap GPUs Have a Local AI Catch Nobody Mentions
• Your OS is quietly choosing how good local AI feels
• Your Next Local AI Upgrade Could Already Be Free
• Jev can’t do what these 2 local AI rivals unlocked
• Ollama’s best trick isn’t the chatbot you downloaded
• Qwen is 3x faster on apple, but should you switch?
• Strix halo vs mac studio: the 128GB buying trap
• Your local AI laptop choice hides a brutal 24GB trap
• Your local coding AI can edit the wrong files confidently
• AI chips hit 17,000 tok/s. Should you ditch GPUs?
• Jev’s big idea makes local qwen do less and act faster
• 8 local AI myths that make you buy the wrong upgrade
• Local AI is free until you see this hardware bill
• The Four Bit Setting That Makes Huge AI Models Fit
• Your 128K local AI setting has a hidden memory bill
• Your Local AI Is Slow for a Reason You Can Fix
• Your PC Could Run Better Local AI Than You Think
• Your local AI choice could make it feel broken
• 9 Local AI models ranked: the biggest wastes your VRAM
• A $300 PC can run 27B AI, but there’s a huge catch
• Apple put a trillion parameter AI on just 4 macs
• GPT 6 sol’s 91% cost saving puts claude on notice
• Qwen 3.8 flash passed 125 tok/s with this speed trick
• Qwen wants 10 trillion. What happens to local AI?
• The m5 ultra review result nvidia buyers should see
• Your 12GB GPU can run AI you might be overlooking
• More VRAM or faster VRAM? The local AI buying trap
• The largest local AI model is bigger than you think
• Bonsai 2 shrinks local AI to 5.9 GB. What survives?
• Jev claims 200x faster AI and the demo is ridiculous
• Mac mini vs RTX 5090: don’t buy the wrong AI machine
• Deepseek 552B just ran on one mini PC. Here is how
• Your $1,000 local AI budget can buy the wrong machine
• Stop Spending $1000 On The Wrong Local AI Machine
• The best coding AI for 4GB VRAM has a hidden rival
• ChatGPT vs Claude: Your Best Model May Cost Extra
• 128GB vs 192GB: The Local AI Upgrade You Might Regret
• DeepSeek V4.1 Flash Makes Claude Earn Your Money
• A 24GB Local Model Beat Claude Where It Should Not
• Qwen 3.8 Max Is The Open Model Moment We Waited For
• The Local AI Mini PC I Would Actually Buy Today
• The Local AI Model That Finally Stops Flinching
• The Mac Mini Local AI Trap Costs More Than You Think
• This $4,000 Robot Exposes The Local AI Trap Now
• Letta Makes Local AI Finally Remember Your Work
• Your Local AI Is Probably Picking The Wrong Model
• Stop Renting AI Before It Gets Much More Expensive
• Your Local AI Setup Needs More VRAM Than You Think
• Nvidia DGX Spark Just Made Local AI Feel Serious
• The Qwen Speed Run That Changes Local AI Forever
• The Best 16GB Coding Model Is Not The Obvious One
• 192GB Just Broke The Local AI Memory Wall Open
• Your Mac Is Not Slow Your Runtime Is The Problem
• Claude Fable 5.1 Has One Number You Cannot Ignore
• FreeToken Just Broke The Local AI VRAM Limit Again
• Nvidia Just Made AI Agents Too Fast To Ignore!
• OpenAI Codex Can Now Work For Twenty Four Hours
• OpenClaw 2.0 Is Coming For Your AI Coding Stack
• Tiny AI PCs Are Running Models They Should Not
• Your 8GB GPU Can Code Locally If You Pick Right
• Do Not Buy A Local AI Box Until You See This Now
• Hugging Face Just Made AI Physical For Only $399
• OpenAI Codex Wants The Most Dangerous Permission
• A 9B Qwen Model Just Added 14.6 SWE Bench Points
• IBM Granite 4.2: Use 3B Before Claude Or GPT 5
• Tencent Hy4 Just Put Qwen And DeepSeek On Notice
• OpenAI Just Put Nvidias AI Chip Monopoly On Notice
• NVIDIA Groq 3 LPX Makes AI Agents Feel Alive Now
• This Open Source AI Map Feels Like A Command Room
• Apple Just Made The Local AI Box Developers Need
• Mistral HUMAIN Is The AI Warning Developers Need
• NVIDIA Just Made Local AI Too Fast To Ignore Now
• Your AI Agent Is Forgetting The Most Expensive Part
• Your Laptop Is Secretly Wasting Local AI Power
• Claude Code Has A New Problem Nobody Is Measuring
• Foxl Makes Every Other AI Agent Feel Backwards
• MCP Just Became The AI Agent Control Layer War
• NVIDIA Just Proved Your AI Model Is Not Enough
• Qwen 3.8 27B Just Made GPT 5.6 Look Wasteful Now
• The Local AI Repo Everyone Is Quietly Moving To
• Vercel Labs Is Quietly Showing The Future Of Code
• 5 Claude Skills Fix The Dumbest Coding Mistakes
• Alibaba Just Made AI Agents Much Harder To Hide
• DeepSeek V4 Flash Makes Claude Look Expensive Now
• This Free Mystery AI Model Should Not Exist Yet
• Unsloth Makes Fine Tuning Feel Completely Unfair
• 15 Google AI Tools That Make Paid Apps Look Dumb
• Stop Paying Claude Code Prices Use Antigravity Now
• This AI Answers The Questions ChatGPT Still Refuses
• Your AI Agent Needs A Tiny Prison Before It Codes
• Grok Bot Turns One Person Into An AI Team Overnight
• Mojo Just Made Python AI Look Painfully Slow Now
• OpenAI Astra Just Made GPT 6 Feel Shockingly Close
• OpenAI’s $12B Bet Just Exposed The Whole AI Bubble
• Someone Ran Kimi K3 On 8 GB RAM And It Hurts Badly
• Your Phone Can Run Real AI Now And That Changes Apps
• Claude Watermarks Already Have A Removal Problem
• DeepSeek Harness: 127K Stars and One Huge Warning
• Kimi K3 Has One Million Tokens And One Huge Trap
• Local AI Hardware: The Spec Nobody Warns You About
• Local AI Prompts: The Mistake Making Models Dumb
• GLM 5.3: The Security Agent Test Starts Right Now
• North Mini Code: The Cheap Worker Model Arrived
• OMG: The Missing Ledger For AI Agent Teams Now
• Qwen3.8 Max: The One Million Token Agent Trap Now
• 10 Things Separate Real AI Coding From Vibe Slop
• Postgres for AI Agents: The Whole Backend Trick
• The Least Fashionable Stack Writes The Best AI Code
• The Most Boring Stacks Make AI Write The Best Code
• Your Frontend State Is Confusing Your Coding Agent
• Your Repo Is Making Your Coding Agent Look Stupid
• AI Finds The Five Files That Explain Any Codebase
• Run Real AI Agents In 2.5GB Of RAM With LFM2.5
• The Cheap AI Model Is The Expensive One For This Job
• The Local AI That Reads Your Screen Without Uploading It
• The Truth About Local AI That Sees And Hears You
• Your AI Coding Bill Is Mostly Files That Did Not Matter
• Your AI Writing Now Carries A Mark You Cannot See
• Your Local Coding Model Is Not Useless, It Is Blind
• GitHub Just Made The AI Coding War Impossible To Fake
• It Runs On Your Laptop And Your Employer Can Read It
• Microsoft Doesn’t Need OpenAI To Win AI Coding
• Your AI Coding Agent Is Quietly Skipping Security
• Local AI Just Got Voice, Vision, Video And Routing
• Meta Just Made Local AI Mainstream With Muse Glimmer
• Nvidia Open Sourced The Thing That Picks Your Model
• The Best Open AI Models Are The Ones Nobody Trusts
• Meta’s Coding Agent Is Cheap Because You Are Paying in Code
• Next.js Is Quietly Becoming an AI Agent Framework
• Stop Blaming The Model, Your AI Coding Setup Is Wrong
• Your $200 AI Coding Plan Is Mostly Paying for Waste
• 10 Free Coding Agents That Replace Paid AI Tools
• GitHub Repos Are Banning AI And Agents Ignore It
• Your CLAUDE.md Is Probably Wrong And Making Claude Worse
• Your Codebase Could Soon Run Itself, Cursor Update
• 10 Claude Skills That Stop You Repeating Yourself
• Open Weights Is Not Open Source And Kimi Proved It
• The Truth About Running AI On Your Own Hardware
• Why AI Agents Are Turning Into Graphs Instead of Loops
• These AI Agents Are Getting Too Good at Hacking
• Buzz Wants To Replace Slack And GitHub Completely
• Forget AI Coders. Agent Swarms Are Here (Run A Dev Team)
• Run A Local AI Model On Zero Video Memory For Free
• The 10 Best Free Local AI Models That Don’t Suck
• Next.js Has an Astro Problem and the Numbers Show It
• Three Ways To Run Kimi K3, And Only One Of Them Is Free
• Why Disney Dropped GitHub Copilot For OpenAI Codex
• $20 of AI a Month. One of These Two Is Wasting It.
• Nobody Should Pay for AI Coding Anymore (Open Models Won)
• 10 Open Source AI Tools That Feel Illegal To Know
• Rust Just Deleted Your Build Step (Vite 8 Is 30x Faster)
• Stop Paying for a Coding Model. Laguna S Costs 20 Cents
• The Best Local AI Model for 8GB of VRAM in 2026
• Ex OpenAI CTO Shipped a 975B Model Free: Can You Run It?
• Open Models Are 14x Cheaper Per Task. Here Is The Catch
• Postgres Destroyed Neo4j. And It’s Free
• Rust Just Deleted Your Build Step (Vite 8 Is 30x Faster)
• The Biggest AI Model Ever Built Just Went Free (Kimi K3)