06 — Air-Gapped Claude Code¶

▤ Summary

This article explains how to run Claude Code locally on Apple Silicon with Ollama and an MLX-backed model so nothing leaves the machine. It walks through the exact setup, including the required version, environment variables, and a smoke test against a Kubernetes cluster. It also details four fixes needed to make the workflow function reliably, such as disabling thinking and upgrading Ollama. The piece shows that hardware, especially memory bandwidth and available unified memory, determines speed and usable context window size. It positions the workflow for teams that need air-gapped operation and local-only handling of cluster data and source code.

▦ Linked Companies