February 2026 - Present
A compute node so time-consumed that I needed another project page for it.
Specs:
Its low power footprint keeps power draw low and can still bring multicore performance.
Using Docker, I deployed OpenWebUI, a web UI for sending and receiving messages with LLMs.
Halo is the name of the agent that connects with my Discord and terminals. After hearing about the ease of having a coding agent in an app I already have and use like WhatsApp or Discord, I decided to set it up. For much of my tuning, this was a great "average use workload" for my LLMs, with context in high demand in a multi-modal interface.
This project demanded the most memory and compute, so it was also the most tuned