Give your computer
a local workshop.
By Alvin
A local AI model runs on your own computer. In this workshop you will ask one to write a small Python program, run the result against invented notes, then change the prompt and inspect the new build.
Three pieces you can name
The model is a downloaded set of learned numbers, called weights. A runner such as Ollama loads it and produces a reply. A coding agent adds tools for reading files, editing code and running checks. The model’s context is the text it can consider in a request; a token is a small piece of that text.
Our experiment asks the model for one Python function: a reusable piece of code that takes notes and returns organized data. The supplied test driver calls that function with four examples and compares the answers with expected results. We call those examples fixtures. Each tested function is fresh model output. The driver removes surrounding code-block backticks and keeps the original reply for inspection.
The earlier small-model measurements are local observations from November 14, with a November 15 rerun. The expanded Edge0 workshop adds a November 17 replay, two fresh programs and behavior checks.