Hosted open-model chat, desktop runners, one local stack, offline tasks, hardware, updates, and sharing a machine with family.
- 1 Easiest path: a hosted open model chat UI If you want to talk to Llama, Qwen, or DeepSeek-class weights without installing a driver, open a hosted chat UI such as Groq, Together, or Fireworks. Spend fifteen minutes on one named model and one public task, then read retention, because the prompt still leaves. Scheduled · September 14, 2026
- 2 Friendly desktop runners from zero Desktop runners like Ollama and LM Studio are how normal people get open weights onto a laptop. Install one app from the official site, pull one small local model, send one public prompt, then find the cloud toggle and the models folder before a 4.1 GB file shows up three times. Scheduled · September 15, 2026
- 3 One local stack end to end (Ollama-style) Kenji had Ollama, LM Studio, and a Docker llama.cpp on three ports, then sent a prompt to the one that was off. This part picks one local stack, walks a 30-minute Ollama path, and shows where the model file and the RAM live. Scheduled · September 16, 2026
- 4 First useful offline tasks Hana had a 6-hour flight, a 7B already on disk, and a 2,400-word draft that still opened with As an AI. Use the local model to rewrite, outline, extract fields, or quiz yourself from your own file, not for live web facts, citations, current prices, or anything that must be true tomorrow. Scheduled · September 17, 2026
- 5 Hardware reality: RAM, GPU, heat, battery Apple Silicon shares one memory pool; a PC splits RAM and VRAM. A 7B Q4 on 16 GB is the honest starter: a GPU speeds small models but is not required, and heat plus battery are the tax, not a 4090 for email. Scheduled · September 18, 2026
- 6 Updating models without breaking your setup Ivy pulled a moving tag on Monday and her 12 canned prompts started rambling on a different 4.7 GB file. Pin a known-good name, test one prompt, then delete the old blob so disk and answers stay put. Scheduled · September 19, 2026
- 7 Sharing a local setup with family or a small team Omar opened Ollama at 9:40 p.m. and found a six-line summary of his partner’s therapy journal in the sidebar, after the kids used the same app. This last part of Run open models from scratch treats a kitchen laptop as a shared brain: chats, files, and who can hit the port. Scheduled · September 20, 2026