Claude Code

Lesson learned from running local AI on limited hardware 💻🤖

recently discovered that it’s possible to run Claude Code locally and offline using Ollama — instantly exciting for private, local AI development.

Then reality kicked in.

My machine: MacBook Air M2 (256GB SSD, 8GB RAM)

Free storage at the time: ~3GB — not nearly enough to experiment with large models.

jkhg

IWhile searching for a workaround, I found a cleanup tool called Mole. Clearing caches, unused files, and system clutter freed up ~22GB — a big win. With enough disk space, I installed Ollama and downloaded gpt-oss:20b (~13GB).

Installation went smoothly.

Running it with Claude Code? 💥 My laptop crashed.

The real bottleneck was clear:

👉 An 8GB RAM MacBook Air cannot handle a 20B parameter model, especially for code-heavy workloads.

Key takeaways:

Storage enables installation — RAM enables execution

Bigger models ≠ better experience on constrained hardware

Local AI is powerful, but model size must match your system

Cleanup tools can unlock possibilities you didn’t know you had

This hands-on failure taught me more than any spec sheet ever could.

Lesson for fellow devs experimenting with local AI:

check your hardware limits before going big 🚀

Resources: