Local AI Foundry
How useful can small, free AI tools become on an ordinary desktop?
The problem
Why this exists
Model demos often hide cost, latency, hardware limits, and privacy behind a polished interface. I wanted a place where those tradeoffs stayed visible.
How it works
A small system, end to end
- 01
Runs a prompt through an owner-selected set of local models one at a time and records speed, load time, and memory use.
- 02
Compares Windows, Piper, and Kokoro speech voices through one consistent listening experience.
- 03
Keeps models and generated audio on the desktop, with phone access limited to a private network.
What I learned
A smaller model that starts quickly and behaves predictably can be more useful than a larger model that technically runs. Measuring the waiting is part of measuring the model.
What’s next
Turn the best local capabilities into small, reusable tools instead of treating the lab as a benchmark museum.