Today I reached the limit of what my local machine can handle with my Zephyr bot project. I stretched the system until it cracked, and realized it's time for some serious upgrades.
I spent today editing and updating the personality features for my Hugging Face-based Zephyr bot. I added logging functions and tried introducing more customized behavior into app.py. However, the 16GB memory of my Hugging Face free tier quickly showed its limits.
Every new tweak and addition pushed the system harder. After a few iterations, my laptop — already strained — simply couldn’t handle it. I attempted to install and run phi-4 locally, and it essentially crashed the machine. Opening Blender? Forget about it. Running a personal LLM? Out of reach for now.
It's clear that if I want to continue building my own AI systems and running models locally, I need stronger hardware. I'm currently looking at starting with an M4 Mac Mini with at least 16GB RAM. This would give me enough baseline performance to continue growing my skills without hitting a memory wall every few steps.
Long term, I want to build a full customized system — designed specifically for AI development, 3D rendering, and local LLM hosting. I imagine an expandable workstation where I can host my bots, build my art, and continue crafting every part of the system.
Hugging Face free tier has been an excellent playground. It showed me where my limits currently are, but more importantly, it showed me where I want to go. Pushing Zephyr bot to its current limits taught me a lot about architecture, memory usage, and the reality of running AI models in constrained environments.
I will continue tweaking app.py and refining the experience. In the meantime, it’s time to build a stronger foundation for the future.
# Sample startup
huggingface-cli login
python app.py
# Sample install attempt for phi-4
pip install phi-4
# Blender crash confirmed
open -a Blender
Back to the main terminal: Arynwood DevLog