Build log

How it started: a dusty GPU and 90 days to fill

A GTX 1080 out of the closet, Windows 11 Pro, Ollama, and Llama 3 8B. The first real build log.

Story 2: How do I control this?

Docker, Tailscale, a firewall, and an 8 GB card. The second real build log.

Story 3: The break

OpenClaw never fully ran. The name that stuck was IronClaw. The third real build log.

Story 4: The pain

A WSL fix wiped the containers. A 16 GB card and a fresh Windows install came after. The fourth real build log.

Story 5: Build, build, build

GPT-OSS 20B on 16 GB, a switch from Claude to GPT, and the lesson that the build was the problem. The fifth real build log.

Story 6: The personal assistant bots

Dots on the $100 plan, a return to $20, and Grok Bot building this site. The sixth real build log.

A 20B model behind a fail-closed wrapper

I let a 20B open-weight model call tools, then took the loop away after two identical calls. The prompt did not do that job.

agents

Q3 was confident and wrong

A 14B at Q3 was quick and calm, and it missed a third of a tiny fixture that Q5 answered.

quality

The context window was a suggestion

The model card said 32k. A second flag had capped the prompt, and the model invented a task I did not ask for.

The card that reported zero

After a routine update the sample bench booted, the desktop came up, and the card reported 0 MB.

hardware