Twenty years of Linux, and now I run my own AI

Short summary

Developer shares 20-year Linux experience running open-source AI models (Gemma, Nemotron, DeepSeek) locally on Apple Silicon via oMLX and Ollama, achieving 27-158 tokens/sec depending on model architecture. Sets up launchd scheduling to automate daily planning, to-do lists, and system checks without consuming cloud API tokens. Avoids high token costs and licensing tiers while gaining freedom to run models continuously and experiment without limits.

• •Local open-source models run efficiently on Apple Silicon (27-158 tokens/sec depending on model architecture like mixture-of-experts)
• •Automate scheduled tasks and continuous processes via launchd without cloud API costs or token consumption limits
• •Reduce spending on cloud APIs while maintaining full control and freedom to experiment with models

Generated with AI, which can make mistakes.