Back to Projects
G3T
Bringing the Cutting Edge of AI to 2005
Overview
There is something magical about solving a problem for a small, limited box.
I recently picked up a PowerMac G3, running Mac OSX 10.4.11 from '99, and I wanted to see how far I can make a system like this usable in the modern day.
I wanted to see if I can design and create a ChatGPT and Ollama client that manages state, sessions, and a local DB.
Bottom Line
Mac G3T is a fully fledged LLM Client, built specifically for legacy systems on Java 5.
This opens the door to more possibilities to use AI to improve and build things fully on piece of computing history.
Tech
- Java 8 (transpiled to Java 5)
- Swing UI
- Retrolambda for backporting lambda support
- HSQLDB + Hibernate
- Bash Scripts for Building and Deployment
What It Does
- I can use my Powermac to chat with my self hosted AI models, and I am currently working on a larger project to bring a TLS 1.2 Proxy to Tiger to use other APIs.
- This allows me to use a piece of history while also enjoying some of the modern wonders of development.
- It is a good testing ground for me to start with, next up I want to make a JellyFin client for Tiger.
Key Decisions
- I wanted to make this compatible with all Macs from the era, and display nicely on them as well. As a result, the default window layout is optimized for the smallest screen that 10.4 systems shipped with.
- Memory is extremely limited on 10.4 - I had to make some shortcuts to the rendering of elements in the lists to avoid excessive memory usage.
- Java 5 was the natural target - all macs from the 10.4 era support it.
Numbers
- 3 scripts - to build, compile, and run it.
- 1 app bundle - easy to install and run for average user.
Status
WIP. Token streaming works fully, and Ollama is planned to be supported first. Next up is small state management things, relating to selecting providers, models, and settings.