Local LLMs and AI agents, built in my home lab

I'm learning AI in the open. I run local models and agents in my home lab and share the work as I go.

Welcome to LLMBits

Hinton Says AI Already Thinks. I Read Its Chain of Thought

public – 9 min read

–

Geoffrey Hinton told StarTalk that language models already think, and that you can watch them do it. I ran his captain puzzle and a shop math problem through three small reasoning models in my home lab, thinking on and off, and read every trace.

Oct 23, 2026

How Claude Duet Works, and the Home Lab Version

public – 3 min read

–

Anthropic opened a research preview of Claude Duet, a proactive agent. Here's what it documents about triggers, background execution, memory, tools and approvals, what's still unknown, and how the same pieces map to Ollama and Hermes in my home lab.

Oct 21, 2026

Run Hermes Agent with Local Ollama on Ubuntu

public – 8 min read

–

I installed Nous Research's Hermes Agent next to Ollama on a CPU-only Ubuntu VM. One 1.9 GB Qwen3.5 build ran real tools. Six smaller-model attempts never ran a tool at all.

Oct 21, 2026

Tiny Ollama Models on an Old Xeon: CPU Shootout

public – 9 min read

–

Gemma 3 1B, Llama 3.2 1B, Qwen3 0.6B, SmolLM2 360M and Granite 3.3 2B on a CPU-only Xeon E5-2670 VM. 60 real runs: tokens per second, RAM, and which ones actually got the answers right.

Oct 21, 2026

Crawl4AI and Ollama: Web Pages to Local Summaries

public – 8 min read

–

I ran Crawl4AI 0.9.2 in Docker, crawled one of my own posts into clean Markdown, and had Gemma 3 1B summarize it on CPU. 53 seconds end to end, and a summary that needed fact-checking.

Oct 21, 2026

LibreChat + Ollama: Local AI Chat in My Home Lab

public – 8 min read

–

LibreChat next to Open WebUI on the same CPU-only Ollama box: the clone, the config files, two startup fixes, a real two-turn chat, and what the five-container stack actually costs in RAM.

Oct 21, 2026

Open WebUI + Ollama: Local Chat on Ubuntu

public – 8 min read

–

I gave my CPU-only Ollama install a real chat UI with Open WebUI in Docker. Real commands, real screenshots, one 'does not support tools' error, and two confidently wrong answers.

Oct 21, 2026

Install Ollama on Ubuntu and Run Gemma 3 1B on CPU

public – 5 min read

–

One install command, one 806 MB model, and 25 tokens/s on CPU. Here's exactly what I ran on an Ubuntu VM built in my home lab, plus the two gotchas that tripped me up.

Oct 20, 2026