journal entry
Why I'm Building a Local AI Lab
Notes on learning local LLMs, hardware limits, self-hosting, and practical AI workflows.
Local AI is interesting because it turns abstract AI talk into practical systems work.
What can run on the hardware I actually have? Which models are useful enough to keep? What does latency feel like in a real workflow? What breaks when the machine is under load? Which prompts become reusable instead of clever once?
Those are homelab questions as much as AI questions.
Things to track
- Hardware, memory, thermals, storage, and power draw.
- Ollama and Open WebUI setup notes.
- Models that feel useful for specific tasks.
- Prompt patterns worth saving.
- Retrieval or document workflows that survive real use.
- Where cloud AI is still the better tool.
The goal is not to pretend every workload should be local. The goal is to understand what local AI is good at by using it in a real environment.