A private ChatGPT for the whole house on a PC with two graphics cards from 2016. Every command, one custom model per person with their own instructions, the two house lines that fix the date and the ‘where do my questions go’ answer, the one setting that stops everyone waiting in line, and the three-day Windows default that killed ours.
We took a stock Asus TUF gaming laptop with an RTX 3070 and 8 GB of video memory, and benched eight local models overnight across ten tasks. The result that surprised us: a 30 billion parameter model with two thirds of itself off the card beat a 14 billion parameter model that fit better, by nearly four times. File size does not predict speed. All 103 measured rows are free below, failures included.
Three open-weight models, one RTX 5090, and the same jobs: a real bug, playable games, tool calling, long documents. Measured numbers, two defaults that were quietly breaking the results, and the finding we published wrong and had to correct.
The bridge nobody documents: connect a local Ollama brain to a real-time 3D VRM face with voice, lip-sync and expressions. The four services, the ports they talk on, and the two traps that cost us days, the thinking-mode stall and the open-mic mistake.
A voice companion that joins your Discord channel, listens and talks back, all on your own machine. Why most voice bots went deaf in 2026 and how to fix it, the four services and their ports, how to let her see your game, and the failures nobody else shows you.
A local agent becomes genuinely useful the day you can teach it: skills are small text files it reads before acting. Here are the two we taught ours this week, summarize any YouTube link, answer only from my own notes, both downloadable, plus the GPU tax nobody warns you about and the honest bill.
A local model can join your Minecraft world and actually play, gather, craft, follow you, even recover from its own mistakes. Here is the working setup, the two silent failures that will waste your evening if nobody warns you, and an honest account of what an 8B model can and cannot do in there.
8 million people left Character.AI, and most ran to another cloud that does the same thing. Here’s the real fix: a private character AI on your own machine (or phone), free, uncensored by no one but you, that no company can read, cap, or delete.
Week of 20 July: open-source AI went enormous, Qwen 3.8 at 2.4 trillion parameters, Kimi K3 at 2.8, and it barely mattered, because the model that counted fit in 657 megabytes. Plus GLM-5.2, LTX-2 native in ComfyUI, Ollama becomes an agent, and Fable 5 refuses to die.
One sentence in, lettered comic pages out, a local LLM writes the script, your GPU draws every panel with the same character in the same style, and an open-source compositor adds bubbles, SFX, and diagonal panel cuts. Tool included.